Calling strategies

strategies = Rtrack::call_strategy(experiment$metrics)

Thresholding strategies

limits called strategies to those, where confidence is greater than 40%

dim(Rtrack::threshold_strategies(strategies, 0.4)$calls)

[1] 4203   12

back to top

Plotting strategies of all age groups combined

par(mfrow = c(2, 2))
Rtrack::plot_strategies(strategies, experiment = experiment, factor = "Strain",
    exclude.probe = TRUE)

back to top

Plotting thresholded strategies of all age groups combined

par(mfrow = c(2, 2))
Rtrack::plot_strategies(Rtrack::threshold_strategies(strategies, 0.4), experiment = experiment,
                        factor = "Strain", exclude.probe = TRUE)

back to top

Saving the results

Here we export the results of the analyzed Track Files into a data.frame, to analyse them further.

results = Rtrack::export_results(experiment)
datatable(results, rownames = FALSE, filter="top", options = list(pageLength = 5, scrollX=T) )

back to top

Plotting with ggplot2

library(ggplot2)
library(readxl)
Results=read_excel("Results3.xlsx", col_types=c("text","text","text","text","text","text","logical","numeric","text","text","text","numeric","text","numeric","text","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric","numeric"))
ggplot(Results, aes(x=`_Day`,y=path.length,color=factor(Strain)))+geom_jitter()

Results

# A tibble: 5,960 x 60
   Track_ID `_TargetID` `_Day` `_Trial` `_Arena`    Condition Probe Trial Strain
   <chr>    <chr>       <chr>  <chr>    <chr>       <chr>     <lgl> <dbl> <chr> 
 1 Track_4  60-911      1      1        Arena_SW.t~ WT_STD    FALSE     4 WT    
 2 Track_8  60-293      1      1        Arena_SW.t~ WT_STD    FALSE     8 WT    
 3 Track_9  60-296      1      1        Arena_SW.t~ WT_STD    FALSE     9 WT    
 4 Track_10 60-333      1      1        Arena_SW.t~ dTg_STD   FALSE    10 dTg   
 5 Track_14 60-911      1      2        Arena_SW.t~ WT_STD    FALSE    14 WT    
 6 Track_18 60-293      1      2        Arena_SW.t~ WT_STD    FALSE    18 WT    
 7 Track_19 60-296      1      2        Arena_SW.t~ WT_STD    FALSE    19 WT    
 8 Track_20 60-333      1      2        Arena_SW.t~ dTg_STD   FALSE    20 dTg   
 9 Track_24 60-911      1      3        Arena_SW.t~ WT_STD    FALSE    24 WT    
10 Track_28 60-293      1      3        Arena_SW.t~ WT_STD    FALSE    28 WT    
# ... with 5,950 more rows, and 51 more variables: Housing <chr>,
#   Age_group <chr>, Age_months <dbl>, All <chr>, strategy <dbl>, name <chr>,
#   confidence <dbl>, 1 <dbl>, 2 <dbl>, 3 <dbl>, 4 <dbl>, 5 <dbl>, 6 <dbl>,
#   7 <dbl>, 8 <dbl>, 9 <dbl>, path.length <dbl>, mean.velocity <dbl>,
#   sd.velocity <dbl>, total.time <dbl>, latency.to.goal <dbl>,
#   goal.crossings <dbl>, old.goal.crossings <dbl>, coverage <dbl>,
#   mean.d.centroid <dbl>, mean.d.goal <dbl>, mean.d.old.goal <dbl>, ...

back to top

CAREFUL!!! The graphs below still include the probe trials (day 7 trial 1). To remove them add: Probe==‘FALSE’ to filter

Path Length Graphs

WT_STD=filter(Results, Strain=='WT'&Housing=='STD')
dTg_STD=filter(Results, Strain=='dTg'&Housing=='STD')
APP_STD=filter(Results, Strain=='APPswe'&Housing=='STD')
PS_STD=filter(Results, Strain=='PS1dE9'&Housing=='STD')
WT_ENR=filter(Results, Strain=='WT'&Housing=='ENR')
dTg_ENR=filter(Results, Strain=='dTg'&Housing=='ENR')
APP_ENR=filter(Results, Strain=='APPswe'&Housing=='ENR')
PS_ENR=filter(Results, Strain=='PS1dE9'&Housing=='ENR')
par(mfrow = c(2, 4))

WT_STD %>%
  mutate(WT_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
  ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
    labs(x="Day",
       y="Average Path length",
       title="Mean Path length WT STD")+scale_fill_jco()

dTg_STD  %>%
mutate(dTg_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length dTg STD")+scale_fill_jco()

APP_STD  %>%
mutate(APP_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length APPswe1 STD")+scale_fill_jco()

PS_STD  %>%
mutate(PS_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length PS1dE9 STD")+scale_fill_jco()

WT_ENR  %>%
mutate(WT_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length WT ENR")+scale_fill_jco()

dTg_ENR  %>%
mutate(dTg_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length dTg ENR")+scale_fill_jco()

APP_ENR  %>%
mutate(APP_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length APPswe1 ENR")+scale_fill_jco()

PS_ENR  %>%
mutate(PS_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length PS1dE9 ENR")+scale_fill_jco()+
  facet_wrap(~Age_group)

back to top

Latency Graphs

WT_STD %>%
  mutate(WT_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
  ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
    labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s WT STD")+scale_fill_jco()

dTg_STD  %>%
mutate(dTg_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s dTg STD")+scale_fill_jco()

APP_STD  %>%
mutate(APP_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s APPswe1 STD")+scale_fill_jco()

PS_STD  %>%
mutate(PS_STD, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s PS1dE9 STD")+scale_fill_jco()

WT_ENR  %>%
mutate(WT_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s WT ENR")+scale_fill_jco()

dTg_ENR  %>%
mutate(dTg_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s dTg ENR")+scale_fill_jco()

APP_ENR  %>%
mutate(APP_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s APPswe1 ENR")+scale_fill_jco()

PS_ENR  %>%
mutate(PS_ENR, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, fill=Age_group))+geom_boxplot()+
  labs(x="Day",
       y="Average Latency in s",
       title="Mean Latency in s PS1dE9 ENR")+scale_fill_jco()+
  facet_wrap(~Age_group)

back to top

Heatmap_probe

par(mfrow = c(2, 2))
Rtrack::plot_density(wtp.metrics, title = "WT Probe Heatmap 3-25mo",
                     col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgp.metrics, title = "dTg Probe Heatmap 3-25mo",
                     col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(APPp.metrics, title = "APPswe Probe Heatmap 3-25mo",
                     col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(PS1p.metrics, title = "PS1dE9 Probe Heatmap 3-25mo",
                     col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

par(mfrow = c(2, 4))
Rtrack::plot_density(wtpstd.metrics, title = "WT Probe Heatmap STD 3-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpstd.metrics, title = "dTg Probe Heatmap STD 3-25mo",    col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(APPpstd.metrics, title = "APPswe Probe Heatmap STD 3-25mo", col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(PS1pstd.metrics, title = "PS1dE9 Probe Heatmap STD 3-25mo", col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(wtpenr.metrics, title = "WT Probe Heatmap ENR 3-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpenr.metrics, title = "dTg Probe Heatmap ENR 3-25mo",    col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(APPpenr.metrics, title = "APPswe Probe Heatmap ENR 3-25mo", col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(PS1penr.metrics, title = "PS1dE9 Probe Heatmap ENR 3-25mo", col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

WT -> dTg -> APPswe -> PS1dE9

back to top

par(mfrow = c(1, 3))
###########WT_Probe##########
Rtrack::plot_density(wtpstd3.metrics, title = "WT Probe Heatmap STD 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(wtpstd14.metrics, title = "WT Probe Heatmap STD 13-14mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(wtpstd25.metrics, title = "WT Probe Heatmap STD 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

Rtrack::plot_density(wtpenr3.metrics, title = "WT Probe Heatmap ENR 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(wtpenr14.metrics, title = "WT Probe Heatmap ENR 13-14mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(wtpenr25.metrics, title = "WT Probe Heatmap ENR 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

###########dTg_Probe##########
Rtrack::plot_density(dTgpstd3.metrics, title = "dTg Probe Heatmap STD 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpstd14.metrics, title = "dTg Probe Heatmap STD 13-14mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpstd25.metrics, title = "dTg Probe Heatmap STD 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

Rtrack::plot_density(dTgpenr3.metrics, title = "dTg Probe Heatmap ENR 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpenr14.metrics, title = "dTg Probe Heatmap ENR 13-14mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(dTgpenr25.metrics, title = "dTg Probe Heatmap ENR 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

###########APP_Probe##########
Rtrack::plot_density(APPpstd3.metrics, title = "APPswe Probe Heatmap STD 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(APPpstd25.metrics, title = "APPswe Probe Heatmap STD 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(APPpenr3.metrics, title = "APPswe Probe Heatmap ENR 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

Rtrack::plot_density(APPpenr25.metrics, title = "APPswe Probe Heatmap ENR 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
###########PS1_Probe##########
Rtrack::plot_density(PS1pstd3.metrics, title = "PS1dE9 Probe Heatmap STD 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(PS1pstd25.metrics, title = "PS1dE9 Probe Heatmap STD 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

Rtrack::plot_density(PS1penr3.metrics, title = "PS1dE9 Probe Heatmap ENR 3mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))
Rtrack::plot_density(PS1penr25.metrics, title = "PS1dE9 Probe Heatmap ENR 17-25mo",
  col = colorRampPalette(c("#000C9E", "#00FEF6", "#FE009E"))(100))

back to top

Path Length Graphs

mo3=filter(Results, Age_group=='3')
mo14=filter(Results, Age_group=='13-14')
mo25=filter(Results, Age_group=='17-25')

mo3 %>%
  mutate(mo3, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
  ggplot(aes(x=`_Day`,y=path.length, fill=Condition))+geom_boxplot()+
    labs(x="Day",
       y="Average Path length",
       title="Mean Path length 3mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#ad5fc9", "dTg_STD" = "#bc91cc","WT_ENR" = "#6eca64", "WT_STD" = "#98cc93","APPswe_ENR" = "#d6564b", "APPswe_STD" = "#db867f","PS1dE9_ENR" = "#918730", "PS1dE9_STD" = "#c5b740"))

mo14  %>%
mutate(mo14, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Condition))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length 13-14mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#ad5fc9", "dTg_STD" = "#bc91cc","WT_ENR" = "#6eca64", "WT_STD" = "#98cc93","APPswe_ENR" = "#d6564b", "APPswe_STD" = "#db867f","PS1dE9_ENR" = "#918730", "PS1dE9_STD" = "#c5b740"))

mo25  %>%
mutate(mo25, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, fill=Condition))+geom_boxplot()+
  labs(x="Day",
       y="Average Path length",
       title="Mean Path length 17-25mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#6c7ed7", "dTg_STD" = "#909ef3","WT_ENR" = "#9f48a3", "WT_STD" = "#ce73cf","APPswe_ENR" = "#c85632", "APPswe_STD" = "#e9724b","PS1dE9_ENR" = "#9f9201", "PS1dE9_STD" = "#cab95b"))

back to top

Boxplot explanation:
Middle line in box -> Median
Box -> shows middle 50% of data(Distance between 1. and 3. Quartil)
Whisker(vertikal lines) -> show upper/lower 25% of data w/o outliers
Points -> outlier

Velocity (Mean) Graphs

mo3=filter(Results, Age_group=='3')
mo14=filter(Results, Age_group=='13-14')
mo25=filter(Results, Age_group=='17-25')

mo3 %>%
  mutate(mo3, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
  ggplot(aes(x=`_Day`,y=mean.velocity, fill=Condition))+geom_boxplot()+
    labs(x="Day",
       y="Average Velocity",
       title="Mean Velocity 3mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#ad5fc9", "dTg_STD" = "#bc91cc","WT_ENR" = "#6eca64", "WT_STD" = "#98cc93","APPswe_ENR" = "#d6564b", "APPswe_STD" = "#db867f","PS1dE9_ENR" = "#918730", "PS1dE9_STD" = "#c5b740"))

mo14  %>%
mutate(mo14, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=mean.velocity, fill=Condition))+geom_boxplot()+
  labs(x="Day",
       y="Average Velocity",
       title="Mean Velocity 13-14mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#ad5fc9", "dTg_STD" = "#bc91cc","WT_ENR" = "#6eca64", "WT_STD" = "#98cc93","APPswe_ENR" = "#d6564b", "APPswe_STD" = "#db867f","PS1dE9_ENR" = "#918730", "PS1dE9_STD" = "#c5b740"))

mo25  %>%
mutate(mo25, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=mean.velocity, fill=Condition))+geom_boxplot()+
  labs(x="Day",
       y="Average Velocity",
       title="Mean Velocity 17-25mo mice")+scale_fill_manual(values = c("dTg_ENR" = "#6c7ed7", "dTg_STD" = "#909ef3","WT_ENR" = "#9f48a3", "WT_STD" = "#ce73cf","APPswe_ENR" = "#c85632", "APPswe_STD" = "#e9724b","PS1dE9_ENR" = "#9f9201", "PS1dE9_STD" = "#cab95b"))

back to top

Strategy (Thresholded) Graphs

par(mfrow = c(2, 2))
Rtrack::plot_strategies(Rtrack::threshold_strategies(strategies, 0.4), experiment = experiment,
                        factor = "All", exclude.probe = TRUE)

back to top

Probe Trial Number of Goals Crossings

mo3.probe=filter(Results, Age_group=='3'&`_Day`=='7'&`_Trial`=='1')
mo14.probe=filter(Results, Age_group=='13-14'&`_Day`=='7'&`_Trial`=='1')
mo25.probe=filter(Results, Age_group=='17-25'&`_Day`=='7'&`_Trial`=='1')

mo3.probe %>%
  ggplot(aes(x=`Strain`,y=goal.crossings, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Number of Goal Crossings",
       title="Probe Trial Number of Goal Crossings 3mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo14.probe  %>%
ggplot(aes(x=`Strain`,y=goal.crossings, fill=Housing))+geom_boxplot()+
  labs(x="Condition",
       y="Number of Goal Crossings",
       title="Probe Trial Number of Goal Crossings 13-14mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo25.probe  %>%
ggplot(aes(x=`Strain`,y=goal.crossings, fill=Housing))+geom_boxplot()+
  labs(x="Condition",
       y="Number of Goal Crossings",
       title="Probe Trial Number of Goal Crossings 17-25mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

back to top

Probe Trial Number of former Goals Crossings

mo3.probe=filter(Results, Age_group=='3'&`_Day`=='7'&`_Trial`=='1')
mo14.probe=filter(Results, Age_group=='13-14'&`_Day`=='7'&`_Trial`=='1')
mo25.probe=filter(Results, Age_group=='17-25'&`_Day`=='7'&`_Trial`=='1')

mo3.probe %>%
  ggplot(aes(x=`Strain`,y=old.goal.crossings, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Number of old Goal Crossings",
       title="Probe Trial Number of former Goal Crossings 3mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo14.probe  %>%
  ggplot(aes(x=`Strain`,y=old.goal.crossings, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Number of old Goal Crossings",
       title="Probe Trial Number of former Goal Crossings 13-14mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo25.probe  %>%
  ggplot(aes(x=`Strain`,y=old.goal.crossings, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Number of Old Goal Crossings",
       title="Probe Trial Number of former Goal Crossings 17-25mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

back to top

Probe Trial Time in old goal zone

mo3.probe=filter(Results, Age_group=='3'&`_Day`=='7'&`_Trial`=='1')
mo14.probe=filter(Results, Age_group=='13-14'&`_Day`=='7'&`_Trial`=='1')
mo25.probe=filter(Results, Age_group=='17-25'&`_Day`=='7'&`_Trial`=='1')

mo3.probe %>%
  ggplot(aes(x=`Strain`,y=time.in.zone.old.goal, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Time in s",
       title="Probe Trial Time spent in old goal 3mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo14.probe  %>%
  ggplot(aes(x=`Strain`,y=time.in.zone.old.goal, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Time in s",
       title="Probe Trial Time spent in old goal 13-14mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo25.probe  %>%
  ggplot(aes(x=`Strain`,y=time.in.zone.old.goal, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Time in s",
       title="Probe Trial Time spent in old goal 17-25mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

back to top

Probe Trial Mean Initial heading error

mo3.probe=filter(Results, Age_group=='3'&`_Day`=='7'&`_Trial`=='1')
mo14.probe=filter(Results, Age_group=='13-14'&`_Day`=='7'&`_Trial`=='1')
mo25.probe=filter(Results, Age_group=='17-25'&`_Day`=='7'&`_Trial`=='1')

mo3.probe %>%
  ggplot(aes(x=`Strain`,y=`mean.initial.heading.error`, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Initial heading error in %??",
       title="Probe Trial Mean Initial heading error 3mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo14.probe  %>%
  ggplot(aes(x=`Strain`,y=`mean.initial.heading.error`, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Initial heading error in %??",
       title="Probe Trial Mean Initial heading error 13-14mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

mo25.probe  %>%
  ggplot(aes(x=`Strain`,y=`mean.initial.heading.error`, fill=Housing))+geom_boxplot()+
    labs(x="Condition",
       y="Initial heading error in ???",
       title="Probe Trial Mean Initial heading error 17-25mo mice")+scale_fill_manual(values = c("ENR"="#b3669e","STD"="#98984d"))

back to top

Strategy Legend

• 1 - thigmotaxis
  
• 2 - circling
  
• 3 - random path
  
• 4 - scanning
  
• 5 - chaining
  
• 6 - directed search
  
• 7 - corrected search
  
• 8 - direct path
  
• 9 - perseverance

Strategies 1-5 are less hippocampus-dependent, ‘unspatial’ and therefore get strategy.class = 0. In contrast to that, strategies 6-9 are more hippocampus-dependent, ‘spatial’ and get strategy.class = 1 assigned.

All probabilities in the following codes relate to choosing strategy.class = 1.

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
logistic.data=read_excel("logistic.reg.xlsx")

Age group 17-25 Housing

logistic.data.1725=filter(logistic.data, Age_group1725=='1')

fitting different logistic regression models with increasing complexity
simple model including housing only

logreg.1725.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.1725)

simple model including Strain only

logreg.1725.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.1725)

additive model including all factors

logreg.1725.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.1725)

interaction model including all factors

logreg.1725.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.1725)

Model comparisons using the anova function

anova(logreg.1725.add.1b, logreg.1725.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      1308     1775.6                          
2      1310     1811.7 -2  -36.096 1.452e-08 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(logreg.1725.add.2, logreg.1725.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1      1307     1774.6                     
2      1308     1775.6 -1  -1.0162   0.3134

adding factor housing does not improve the model

anova(logreg.1725.int, logreg.1725.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1      1304     1771.3                     
2      1308     1775.6 -4  -4.3588   0.3596

interaction model not better than additive model simple model including Strain only is the minimal adequate model

summary of minimal adequate model

summary(logreg.1725.add.1b)

Call:
glm(formula = strategy.class ~ Strain, family = binomial, data = logistic.data.1725)

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-1.325  -1.222   1.037   1.037   1.475  

Coefficients:
             Estimate Std. Error z value Pr(>|z|)    
(Intercept)   0.10318    0.14378   0.718 0.472979    
StraindTg    -0.78083    0.20945  -3.728 0.000193 ***
StrainPS1dE9 -0.03339    0.20966  -0.159 0.873472    
StrainWT      0.23739    0.16168   1.468 0.142040    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1813.9  on 1311  degrees of freedom
Residual deviance: 1775.6  on 1308  degrees of freedom
AIC: 1783.6

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1725.add.1b))

 (Intercept)    StraindTg StrainPS1dE9     StrainWT 
   1.1086957    0.4580270    0.9671628    1.2679321 

impact of predictor “Strain”

pred.data <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pred.data$prob <- predict(logreg.1725.add.1b, newdata = pred.data, type = "response")
pred.data

  Strain      prob
1     WT 0.5843293
2    dTg 0.3367876
3 PS1dE9 0.5174419
4 APPswe 0.5257732

impact of predictor “housing”

pred.data2 <- data.frame(Housing=c("STD","ENR"))
pred.data2$prob <- predict(logreg.1725.add.1a, newdata = pred.data2, type = "response")
pred.data2

  Housing      prob
1     STD 0.5504451
2     ENR 0.5094044

Age group 13-14 Strain x Housing effects

logistic.data.1314=filter(logistic.data, Age_group1314=='1')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.1314.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.1314)

simple model including Strain only

logreg.1314.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.1314)

additive model including all factors

logreg.1314.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.1314)

interaction model including all factors

logreg.1314.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.1314)

model comparisons using the anova function

anova(logreg.1314.add.1b, logreg.1314.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1      1306     1724.2                     
2      1306     1804.4  0   -80.16         

no Df, same model fit?

anova(logreg.1314.add.2, logreg.1314.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1      1305     1718.1                       
2      1306     1724.2 -1  -6.1765  0.01295 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

additive model shows significantly better fit (Pr=0.01295)

anova(logreg.1314.int, logreg.1314.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)   
1      1304     1712.2                        
2      1306     1724.2 -2  -11.999  0.00248 **
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than simple Strain model (Pr=0.00248) interaction model including Strain*Housing is the best fitting model

summary of minimal adequate model

summary(logreg.1314.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.4842  -0.9285  -0.9253   1.1143   1.4525  

Coefficients:
                     Estimate Std. Error z value Pr(>|z|)    
(Intercept)         -0.626772   0.118643  -5.283 1.27e-07 ***
HousingSTD           0.008426   0.167152   0.050   0.9598    
StrainWT             1.324112   0.163290   8.109 5.11e-16 ***
HousingSTD:StrainWT -0.555484   0.230430  -2.411   0.0159 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1811.7  on 1307  degrees of freedom
Residual deviance: 1712.3  on 1304  degrees of freedom
AIC: 1720.3

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1314.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          0.5343137           1.0084617           3.7588467           0.5737945 

impact of predictor “Strain*Housing”

pred.data1314 <- data.frame(Strain=c("WT", "WT","dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pred.data1314$prob <- predict(logreg.1314.int, newdata = pred.data1314, type = "response")
pred.data1314

  Strain Housing      prob
1     WT     STD 0.5375000
2     WT     ENR 0.6675978
3    dTg     STD 0.3501577
4    dTg     ENR 0.3482428

Age group 3 Strain x Housing

logistic.data.3=filter(logistic.data, Age_group3=='1')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.3.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.3)

simple model including Strain only

logreg.3.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.3)

additive model including all factors

logreg.3.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.3)

interaction model including all factors

logreg.3.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.3)

comparisons between model using the anova function

anova(logreg.3.add.1b, logreg.3.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      3336     4458.5                          
2      3338     4516.0 -2  -57.513 3.245e-13 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model with Strain only is really good (Pr=3.245e-13)

anova(logreg.3.add.2, logreg.3.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      3335     4440.3                          
2      3336     4458.5 -1  -18.243 1.945e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit (Pr=1.945e-5)

anova(logreg.3.int, logreg.3.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      3332     4428.8                          
2      3336     4458.5 -4  -29.694 5.648e-06 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model (Pr=5.648e-6) interaction model is the best

summary of Strain only model

summary(logreg.3.add.1b)

Call:
glm(formula = strategy.class ~ Strain, family = binomial, data = logistic.data.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.4508  -1.3122   0.9267   1.0483   1.2183  

Coefficients:
             Estimate Std. Error z value Pr(>|z|)    
(Intercept)   0.31144    0.06899   4.514 6.35e-06 ***
StraindTg    -0.40719    0.09715  -4.192 2.77e-05 ***
StrainPS1dE9  0.31166    0.09914   3.143  0.00167 ** 
StrainWT      0.30702    0.10290   2.984  0.00285 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 4529.0  on 3339  degrees of freedom
Residual deviance: 4458.5  on 3336  degrees of freedom
AIC: 4466.5

Number of Fisher Scoring iterations: 4

exp(coef(logreg.3.add.1b))  

 (Intercept)    StraindTg StrainPS1dE9     StrainWT 
   1.3653846    0.6655165    1.3656859    1.3593683 

summary of interaction model

summary(logreg.3.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.5713  -1.2014   0.8293   1.0221   1.2715  

Coefficients:
                         Estimate Std. Error z value Pr(>|z|)    
(Intercept)              0.571375   0.099833   5.723 1.04e-08 ***
HousingSTD              -0.515022   0.139153  -3.701 0.000215 ***
StraindTg               -0.567129   0.135864  -4.174 2.99e-05 ***
StrainPS1dE9             0.319191   0.145466   2.194 0.028217 *  
StrainWT                -0.008040   0.153664  -0.052 0.958274    
HousingSTD:StraindTg     0.292295   0.195814   1.493 0.135512    
HousingSTD:StrainPS1dE9  0.001266   0.200208   0.006 0.994953    
HousingSTD:StrainWT      0.610701   0.207830   2.938 0.003298 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 4529.0  on 3339  degrees of freedom
Residual deviance: 4428.8  on 3332  degrees of freedom
AIC: 4444.8

Number of Fisher Scoring iterations: 4

exp(coef(logreg.3.int))

            (Intercept)              HousingSTD               StraindTg 
              1.7707006               0.5974872               0.5671514 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.3760135               0.9919925               1.3394975 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              1.0012672               1.8417220 

impact of predictor “Strain”

pred.data3 <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pred.data3$prob <- predict(logreg.3.add.1b, newdata = pred.data3, type = "response")
pred.data3

  Strain      prob
1     WT 0.6498674
2    dTg 0.4760793
3 PS1dE9 0.6509217
4 APPswe 0.5772358

impact of predictors “Strain x Housing”

pred.data3x <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pred.data3x$prob <- predict(logreg.3.int, newdata = pred.data3x, type = "response")
pred.data3x

  Strain Housing      prob
1     WT     STD 0.6590389
2     WT     ENR 0.6372240
3    dTg     STD 0.4455959
4    dTg     ENR 0.5010616
5 PS1dE9     STD 0.5931034
6 PS1dE9     ENR 0.7090069
7 APPswe     STD 0.5140845
8 APPswe     ENR 0.6390805

The full overview of the logistic regression results table is here.

Age group 17-25 acquisition Strain

logistic.data.1725.acq=filter(logistic.data.1725, `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity
simple model including housing only

logreg.1725.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.1725.acq)

simple model including Strain only

logreg.1725.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.1725.acq)

additive model including all factors

logreg.1725.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.1725.acq)

interaction model including all factors

logreg.1725.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.1725.acq)

Model comparisons using the anova function

anova(logreg.1725.acq.add.1b, logreg.1725.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       788     1067.9                          
2       790     1088.0 -2  -20.051 4.427e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(logreg.1725.acq.add.2, logreg.1725.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df  Deviance Pr(>Chi)
1       787     1067.9                      
2       788     1067.9 -1 -0.016692   0.8972

adding factor housing does not improve the model

anova(logreg.1725.acq.int, logreg.1725.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       784     1063.0                     
2       788     1067.9 -4  -4.9469   0.2928

summary(logreg.1725.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.1725.acq)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.1632  -1.1575  -0.8672   1.1974   1.8182  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)  
(Intercept)             -0.69315    0.30619  -2.264   0.0236 *
HousingSTD               0.63599    0.38851   1.637   0.1016  
StraindTg               -0.09097    0.40387  -0.225   0.8218  
StrainPS1dE9             0.59471    0.39935   1.489   0.1364  
StrainWT                 0.64619    0.33547   1.926   0.0541 .
HousingSTD:StraindTg    -1.29223    0.59810  -2.161   0.0307 *
HousingSTD:StrainPS1dE9 -0.70460    0.54826  -1.285   0.1987  
HousingSTD:StrainWT     -0.62265    0.43190  -1.442   0.1494  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1088.1  on 791  degrees of freedom
Residual deviance: 1063.0  on 784  degrees of freedom
AIC: 1079

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1725.acq.int))

            (Intercept)              HousingSTD               StraindTg 
              0.5000000               1.8888889               0.9130435 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.8125000               1.9082569               0.2746572 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              0.4943049               0.5365216 

impact of predictor “Strain”

pred.data.17.acq <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pred.data.17.acq$prob <- predict(logreg.1725.acq.add.1b, newdata = pred.data.17.acq, type = "response")
pred.data.17.acq

  Strain      prob
1     WT 0.4900222
2    dTg 0.2631579
3 PS1dE9 0.4678899
4 APPswe 0.4237288

impact of predictor “Housing”

pred.data.17.acq <- data.frame(Housing=c("STD","ENR"))
pred.data.17.acq$prob <- predict(logreg.1725.acq.add.1a, newdata = pred.data.17.acq, type = "response")
pred.data.17.acq

  Housing      prob
1     STD 0.4516129
2     ENR 0.4370180

Age group 17-25 reversal Strain

logistic.data.1725.rev=filter(logistic.data.1725, `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity
simple model including housing only

logreg.1725.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.1725.rev)

simple model including Strain only

logreg.1725.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.1725.rev)

additive model including all factors

logreg.1725.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.1725.rev)

interaction model including all factors

logreg.1725.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.1725.rev)

Model comparisons using the anova function

anova(logreg.1725.rev.add.1b, logreg.1725.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       516     643.08                          
2       518     662.36 -2   -19.28 6.509e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(logreg.1725.rev.add.2, logreg.1725.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       515     641.11                     
2       516     643.08 -1  -1.9658   0.1609

adding factor housing does not improve the model

anova(logreg.1725.rev.int, logreg.1725.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)   
1       512     627.21                        
2       516     643.08 -4  -15.869 0.003201 **
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model

summary of minimal adequate model

summary(logreg.1725.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.1725.rev)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.6567  -1.2049   0.7647   0.8340   1.6033  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)              0.79851    0.40139   1.989 0.046660 *  
HousingSTD              -0.04082    0.50895  -0.080 0.936071    
StraindTg               -1.75992    0.51715  -3.403 0.000666 ***
StrainPS1dE9            -0.73397    0.53877  -1.362 0.173105    
StrainWT                 0.28141    0.44533   0.632 0.527444    
HousingSTD:StraindTg     1.79069    0.71471   2.505 0.012229 *  
HousingSTD:StrainPS1dE9  0.76474    0.73051   1.047 0.295166    
HousingSTD:StrainWT     -0.16186    0.57128  -0.283 0.776929    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 665.61  on 519  degrees of freedom
Residual deviance: 627.21  on 512  degrees of freedom
AIC: 643.21

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1725.rev.int))

            (Intercept)              HousingSTD               StraindTg 
              2.2222222               0.9600000               0.1720588 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              0.4800000               1.3250000               5.9935897 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              2.1484375               0.8505620 

impact of predictor “Strain”

pred.data.17.revs <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pred.data.17.revs$prob <- predict(logreg.1725.rev.int, newdata = pred.data.17.revs, type = "response")
pred.data.17.revs

  Strain Housing      prob
1     WT     STD 0.7062500
2     WT     ENR 0.7464789
3    dTg     STD 0.6875000
4    dTg     ENR 0.2765957
5 PS1dE9     STD 0.6875000
6 PS1dE9     ENR 0.5161290
7 APPswe     STD 0.6808511
8 APPswe     ENR 0.6896552

Age group 13-14 acquisition Strain x Housing

logistic.data.acq.1314=filter(logistic.data.1314,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.1314.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.acq.1314)

simple model including Strain only

logreg.1314.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.acq.1314)

additive model including all factors

logreg.1314.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.acq.1314)

interaction model including all factors

logreg.1314.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.acq.1314)

comparisons between model using the anova function

anova(logreg.1314.acq.add.1b, logreg.1314.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       779     1014.5                     
2       779     1049.4  0  -34.833         

model with Strain only is good

anova(logreg.1314.acq.add.2, logreg.1314.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       778     1009.0                       
2       779     1014.5 -1  -5.5851  0.01811 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit

anova(logreg.1314.acq.int, logreg.1314.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       777     1008.2                       
2       779     1014.5 -2  -6.3566  0.04166 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

additive model better than interaction model additive model is the best

summary of interactive model

summary(logreg.1314.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.acq.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.2994  -1.0987  -0.7931   1.2582   1.6186  

Coefficients:
                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)          -0.7916     0.1583  -5.001 5.70e-07 ***
HousingSTD           -0.2038     0.2278  -0.895    0.371    
StrainWT              1.0738     0.2100   5.112 3.18e-07 ***
HousingSTD:StrainWT  -0.2664     0.3033  -0.878    0.380    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1055.6  on 780  degrees of freedom
Residual deviance: 1008.2  on 777  degrees of freedom
AIC: 1016.2

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1314.acq.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          0.4531250           0.8155922           2.9265367           0.7660990 

impact of predictors “Strain x Housing”

pred.data1314.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pred.data1314.acq$prob <- predict(logreg.1314.acq.add.2, newdata = pred.data1314.acq, type = "response")
pred.data1314.acq

  Strain Housing      prob
1     WT     STD 0.4682013
2     WT     ENR 0.5565671
3    dTg     STD 0.2545257
4    dTg     ENR 0.3273906

Age group 13-14 reversal Strain x Housing

logistic.data.rev.1314=filter(logistic.data.1314,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.1314.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.rev.1314)

simple model including Strain only

logreg.1314.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.rev.1314)

additive model including all factors

logreg.1314.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.rev.1314)

interaction model including all factors

logreg.1314.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.rev.1314)

comparisons between model using the anova function

anova(logreg.1314.rev.add.1b, logreg.1314.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       525     659.45                     
2       525     710.30  0  -50.855         

anova(logreg.1314.rev.add.2, logreg.1314.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       524     658.28                     
2       525     659.45 -1  -1.1655   0.2803

adding factor does not improve model

anova(logreg.1314.rev.int, logreg.1314.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       523     650.44                       
2       525     659.45 -2    -9.01  0.01105 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model interaction model is the best

summary of interaction model

summary(logreg.1314.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.rev.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.8297  -1.1247   0.6444   0.9049   1.3508  

Coefficients:
                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)          -0.3989     0.1810  -2.204  0.02754 *  
HousingSTD            0.2737     0.2533   1.081  0.27974    
StrainWT              1.8652     0.2799   6.664 2.67e-11 ***
HousingSTD:StrainWT  -1.0586     0.3805  -2.783  0.00539 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 711.87  on 526  degrees of freedom
Residual deviance: 650.44  on 523  degrees of freedom
AIC: 658.44

Number of Fisher Scoring iterations: 4

exp(coef(logreg.1314.rev.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          0.6710526           1.3148789           6.4575163           0.3469306 

impact of predictors “Strain x Housing”

pred.data1314.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pred.data1314.rev$prob <- predict(logreg.1314.rev.int, newdata = pred.data1314.rev, type = "response")
pred.data1314.rev

  Strain Housing      prob
1     WT     STD 0.6640625
2     WT     ENR 0.8125000
3    dTg     STD 0.4687500
4    dTg     ENR 0.4015748

Age group 3 acquisition Strain x Housing

logistic.data.acq.3=filter(logistic.data.3,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.3.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.acq.3)

simple model including Strain only

logreg.3.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.acq.3)

additive model including all factors

logreg.3.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.acq.3)

interaction model including all factors

logreg.3.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.acq.3)

comparisons between model using the anova function

anova(logreg.3.acq.add.1b, logreg.3.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      1989     2683.6                          
2      1991     2753.3 -2  -69.785 7.021e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model with Strain only is really good

anova(logreg.3.acq.add.2, logreg.3.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      1988     2668.9                          
2      1989     2683.6 -1  -14.652 0.0001293 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit

anova(logreg.3.acq.int, logreg.3.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      1985     2653.5                          
2      1989     2683.6 -4  -30.063 4.752e-06 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model interaction model is the best

summary of interaction model

summary(logreg.3.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.acq.3)

Deviance Residuals: 
   Min      1Q  Median      3Q     Max  
-1.412  -1.234   0.960   1.089   1.624  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)              0.30111    0.12520   2.405 0.016175 *  
HousingSTD              -0.59571    0.17841  -3.339 0.000841 ***
StraindTg               -0.66816    0.17434  -3.832 0.000127 ***
StrainPS1dE9             0.23441    0.17942   1.307 0.191370    
StrainWT                -0.08980    0.19226  -0.467 0.640462    
HousingSTD:StraindTg    -0.04436    0.26255  -0.169 0.865840    
HousingSTD:StrainPS1dE9  0.19165    0.25270   0.758 0.448197    
HousingSTD:StrainWT      0.89318    0.26355   3.389 0.000701 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 2762.8  on 1992  degrees of freedom
Residual deviance: 2653.5  on 1985  degrees of freedom
AIC: 2669.5

Number of Fisher Scoring iterations: 4

exp(coef(logreg.3.acq.int))

            (Intercept)              HousingSTD               StraindTg 
              1.3513514               0.5511724               0.5126506 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.2641667               0.9141176               0.9566127 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              1.2112495               2.4428887 

impact of predictors “Strain x Housing”

pred.data3.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pred.data3.acq$prob <- predict(logreg.3.acq.int, newdata = pred.data3.acq, type = "response")
pred.data3.acq

  Strain Housing      prob
1     WT     STD 0.6245211
2     WT     ENR 0.5526316
3    dTg     STD 0.2675439
4    dTg     ENR 0.4092527
5 PS1dE9     STD 0.5328185
6 PS1dE9     ENR 0.6307692
7 APPswe     STD 0.4268775
8 APPswe     ENR 0.5747126

Age group 3 reversal Strain x Housing

logistic.data.rev.3=filter(logistic.data.3,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity simple model including housing only

logreg.3.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = logistic.data.rev.3)

simple model including Strain only

logreg.3.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = logistic.data.rev.3)

additive model including all factors

logreg.3.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = logistic.data.rev.3)

interaction model including all factors

logreg.3.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.rev.3)

comparisons between model using the anova function

anova(logreg.3.rev.add.1b, logreg.3.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1      1343     1614.9                     
2      1345     1618.3 -2  -3.3859    0.184

model with Strain only is really good (Pr=3.245e-13)

anova(logreg.3.rev.add.2, logreg.3.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1      1342     1609.6                       
2      1343     1614.9 -1   -5.326  0.02101 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit (Pr=1.945e-5)

anova(logreg.3.rev.int, logreg.3.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)   
1      1339     1598.6                        
2      1343     1614.9 -4  -16.331 0.002606 **
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model (Pr=5.648e-6) interaction model is the best

summary of interaction model

summary(logreg.3.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = logistic.data.rev.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.8720  -1.4233   0.7836   0.8752   0.9500  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)               1.0234     0.1719   5.953 2.63e-09 ***
HousingSTD               -0.4410     0.2339  -1.886   0.0593 .  
StraindTg                -0.4617     0.2287  -2.019   0.0435 *  
StrainPS1dE9              0.5383     0.2643   2.036   0.0417 *  
StrainWT                  0.1501     0.2705   0.555   0.5790    
HousingSTD:StraindTg      0.7387     0.3282   2.251   0.0244 *  
HousingSTD:StrainPS1dE9  -0.3585     0.3481  -1.030   0.3031    
HousingSTD:StrainWT       0.1640     0.3549   0.462   0.6441    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1622.9  on 1346  degrees of freedom
Residual deviance: 1598.6  on 1339  degrees of freedom
AIC: 1614.6

Number of Fisher Scoring iterations: 4

exp(coef(logreg.3.rev.int))

            (Intercept)              HousingSTD               StraindTg 
              2.7826087               0.6433972               0.6302083 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.7130208               1.1619792               2.0931950 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              0.6987137               1.1781761 

impact of predictors “Strain x Housing”

pred.data3.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pred.data3.rev$prob <- predict(logreg.3.rev.int, newdata = pred.data3.rev, type = "response")
pred.data3.rev

  Strain Housing      prob
1     WT     STD 0.7102273
2     WT     ENR 0.7637795
3    dTg     STD 0.7025316
4    dTg     ENR 0.6368421
5 PS1dE9     STD 0.6818182
6 PS1dE9     ENR 0.8265896
7 APPswe     STD 0.6416185
8 APPswe     ENR 0.7356322

Results

Results table can be seen here

2D - Age group 3

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
probs=read_excel("results_logistic_new.xlsx")
myColors = brewer.pal(4,"Set1")
names(myColors)=levels(probs$Housing)
colScale=scale_colour_manual(name="Housing",values=myColors)
probs$`Experiment_Phase`=as.factor(probs$`Experiment_Phase`)
probs3=filter(probs, Age_group=='3' & Model=='STRAIN*HOUSING')

gg = ggplot(probs3, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() + facet_grid(~Experiment_Phase) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 3 months old mice using Strategy class 1")
gg=gg+colScale

ggplotly(p=gg, height=400, width=800)

Strain*Housing interaction model used for predictions.

2D - Age group 13-14

probs13=filter(probs, Age_group=='13-14')

gg = ggplot(probs13, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() + facet_grid(~Experiment_Phase) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 13-14 months old mice using Strategy class 1")
gg=gg+colScale
ggplotly(gg, height=400, width=800)

Strain*Housing interaction model used for predictions of REV and BOTH phases. In the ACQ experiment phase, the additive model (Strain+Housing) shows the best fit. Take a look here

2D - Age group 17-25

probs17=filter(probs, Age_group=='17-25')

gg = ggplot(probs17, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() + facet_grid(~Experiment_Phase) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 17-25 months old mice using Strategy class 1", caption="test")
gg=gg+colScale
ggplotly(gg, height=400, width=800)

Strategy.class ~ Strain and Strategy.class ~ Housing models were used separately for ACQ and Both experiment phases. REV phase has best model fit with the Strategy.class ~ Strain*Housing interaction model.

3D Surface Plot

x=c("WT","dTg","PS1dE9","APPswe")
y=c("STD","ENR","ACQ STD","ACQ ENR", "REV STD", "REV ENR")
z=rbind(
  c(0.1590389,-0.0544041,0.0931034,0.0140845),
  c(0.137224,0.0010616,0.2090069,0.1390805),
  c(0.1245211,-0.2324561,0.0328185,-0.0731225),
  c(0.0526316,-0.0907473,0.1307692,0.0747126),
  c(0.2102273,0.2025316,0.1818182,0.1416185),
  c(0.2637795,0.1368421,0.3265896,0.2356322)
)
z2=rbind(
  c(0,0,0,0),c(0,0,0,0),c(0,0,0,0),c(0,0,0,0),c(0,0,0,0),c(0,0,0,0))
  
ebene= plot_ly(type="surface", x=~x, y=~y, z=~z)
ebene= ebene %>% add_surface(z=~z2)
ebene= ebene %>% layout( title='Probability of Strategy class 1 in 3 months old mice compared to p=0.5')
ebene

Interactive plot created with plotly(Sievert et al., 2021) library.

testing citations(Garthe et al., 2016)

RTrack package used(Overall, 2020)

Strategy Legend

• 1 - thigmotaxis
  
• 2 - circling
  
• 3 - random path
  
• 4 - scanning
  
• 5 - chaining
  
• 6 - directed search
  
• 7 - corrected search
  
• 8 - direct path
  
• 9 - perseverance

Strategies 1-5 are less hippocampus-dependent, ‘unspatial’ and therefore get strategy.class = 0. In contrast to that, strategies 6-9 are more hippocampus-dependent, ‘spatial’ and get strategy.class = 1 assigned.

All probabilities in the following codes relate to choosing strategy.class = 1.

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
ogistic.data=read_excel("logistic_reg_thresh.xlsx")

Age group 17-25 Housing

ogistic.data.1725=filter(ogistic.data, Age_group1725=='1')

fitting different logistic regression models with increasing complexity
simple model including housing only

ogreg.1725.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.1725)

simple model including Strain only

ogreg.1725.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.1725)

additive model including all factors

ogreg.1725.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.1725)

interaction model including all factors

ogreg.1725.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.1725)

Model comparisons using the anova function

anova(ogreg.1725.add.1b, ogreg.1725.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       932     1124.8                          
2       934     1168.0 -2  -43.211 4.139e-10 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(ogreg.1725.add.2, ogreg.1725.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       931     1123.7                     
2       932     1124.8 -1   -1.068   0.3014

adding factor housing does not improve the model

anova(ogreg.1725.int, ogreg.1725.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       928     1118.0                     
2       932     1124.8 -4  -6.7896   0.1474

interaction model not better than additive model simple model including Strain only is the minimal adequate model

summary of minimal adequate model

summary(ogreg.1725.add.1b)

Call:
glm(formula = strategy.class ~ Strain, family = binomial, data = ogistic.data.1725)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.6554  -1.0524   0.7657   0.7657   1.3078  

Coefficients:
             Estimate Std. Error z value Pr(>|z|)    
(Intercept)   0.69315    0.18058   3.838 0.000124 ***
StraindTg    -0.99462    0.25461  -3.906 9.37e-05 ***
StrainPS1dE9  0.03774    0.26570   0.142 0.887047    
StrainWT      0.38380    0.20539   1.869 0.061669 .  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1171.2  on 935  degrees of freedom
Residual deviance: 1124.8  on 932  degrees of freedom
AIC: 1132.8

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.1725.add.1b))

 (Intercept)    StraindTg StrainPS1dE9     StrainWT 
    2.000000     0.369863     1.038462     1.467857 

impact of predictor “Strain”

pre.data <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pre.data$prob <- predict(ogreg.1725.add.1b, newdata = pre.data, type = "response")
pre.data

  Strain      prob
1     WT 0.7459165
2    dTg 0.4251969
3 PS1dE9 0.6750000
4 APPswe 0.6666667

impact of predictor “housing”

pre.data2 <- data.frame(Housing=c("STD","ENR"))
pre.data2$prob <- predict(ogreg.1725.add.1a, newdata = pre.data2, type = "response")
pre.data2

  Housing      prob
1     STD 0.7075665
2     ENR 0.6532438

Age group 13-14 Strain x Housing effects

ogistic.data.1314=filter(ogistic.data, Age_group1314=='1')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.1314.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.1314)

simple model including Strain only

ogreg.1314.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.1314)

additive model including all factors

ogreg.1314.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.1314)

interaction model including all factors

ogreg.1314.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.1314)

model comparisons using the anova function

anova(ogreg.1314.add.1b, ogreg.1314.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       868     1024.4                     
2       868     1117.1  0   -92.65         

no Df, same model fit?

anova(ogreg.1314.add.2, ogreg.1314.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       867     1023.8                     
2       868     1024.4 -1 -0.58952   0.4426

additive model shows significantly better fit (Pr=0.01295)

anova(ogreg.1314.int, ogreg.1314.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       866     1015.5                       
2       868     1024.4 -2  -8.9111  0.01161 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than simple Strain model (Pr=0.00248) interaction model including Strain*Housing is the best fitting model

summary of minimal adequate model

summary(ogreg.1314.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.9169  -1.0996   0.5890   0.7732   1.2573  

Coefficients:
                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)          -0.1859     0.1403  -1.325  0.18514    
HousingSTD            0.2720     0.2030   1.339  0.18041    
StrainWT              1.8496     0.2176   8.501  < 2e-16 ***
HousingSTD:StrainWT  -0.8813     0.3064  -2.877  0.00402 ** 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1118.3  on 869  degrees of freedom
Residual deviance: 1015.5  on 866  degrees of freedom
AIC: 1023.5

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.1314.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          0.8303571           1.3125529           6.3575894           0.4142512 

impact of predictor “Strain*Housing”

pre.data1314 <- data.frame(Strain=c("WT", "WT","dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data1314$prob <- predict(ogreg.1314.int, newdata = pre.data1314, type = "response")
pre.data1314

  Strain Housing      prob
1     WT     STD 0.7416268
2     WT     ENR 0.8407407
3    dTg     STD 0.5215054
4    dTg     ENR 0.4536585

Age group 3 Strain x Housing

ogistic.data.3=filter(ogistic.data, Age_group3=='1')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.3.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.3)

simple model including Strain only

ogreg.3.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.3)

additive model including all factors

ogreg.3.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.3)

interaction model including all factors

ogreg.3.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.3)

comparisons between model using the anova function

anova(ogreg.3.add.1b, ogreg.3.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      2393     2730.5                          
2      2395     2803.6 -2  -73.096 < 2.2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model with Strain only is really good (Pr=3.245e-13)

anova(ogreg.3.add.2, ogreg.3.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1      2392     2726.2                       
2      2393     2730.5 -1  -4.2311  0.03969 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit (Pr=1.945e-5)

anova(ogreg.3.int, ogreg.3.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1      2389     2724.1                     
2      2393     2730.5 -4  -6.3947   0.1715

additive model better than interaction model (Pr=5.648e-6) additive model is the best

summary of Strain only model

summary(ogreg.3.add.1b)

Call:
glm(formula = strategy.class ~ Strain, family = binomial, data = ogistic.data.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.8414  -1.3582   0.7065   0.8176   1.0068  

Coefficients:
             Estimate Std. Error z value Pr(>|z|)    
(Intercept)   0.92408    0.08834  10.461  < 2e-16 ***
StraindTg    -0.50857    0.12243  -4.154 3.27e-05 ***
StrainPS1dE9  0.33659    0.12929   2.603  0.00923 ** 
StrainWT      0.56859    0.14248   3.991 6.59e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 2805.7  on 2396  degrees of freedom
Residual deviance: 2730.5  on 2393  degrees of freedom
AIC: 2738.5

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.3.add.1b))  

 (Intercept)    StraindTg StrainPS1dE9     StrainWT 
   2.5195531    0.6013573    1.4001601    1.7657813 

summary of additive model model

summary(ogreg.3.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.8573  -1.2841   0.6659   0.7928   1.0743  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)               0.9963     0.1194   8.343  < 2e-16 ***
HousingSTD               -0.1629     0.1776  -0.917  0.35906    
StraindTg                -0.4404     0.1665  -2.645  0.00816 ** 
StrainPS1dE9              0.3971     0.1791   2.216  0.02666 *  
StrainWT                  0.4506     0.2058   2.190  0.02854 *  
HousingSTD:StraindTg     -0.1456     0.2462  -0.592  0.55417    
HousingSTD:StrainPS1dE9  -0.1140     0.2595  -0.440  0.66027    
HousingSTD:StrainWT       0.2445     0.2866   0.853  0.39374    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 2805.7  on 2396  degrees of freedom
Residual deviance: 2724.1  on 2389  degrees of freedom
AIC: 2740.1

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.3.int))

            (Intercept)              HousingSTD               StraindTg 
              2.7083333               0.8496756               0.6437870 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.4874725               1.5692308               0.8644712 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              0.8922211               1.2769194 

impact of predictor “Strain”

pre.data3 <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pre.data3$prob <- predict(ogreg.3.add.1b, newdata = pre.data3, type = "response")
pre.data3

  Strain      prob
1     WT 0.8164794
2    dTg 0.6024096
3 PS1dE9 0.7791411
4 APPswe 0.7158730

impact of predictors “Strain + Housing”

pre.data3x <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data3x$prob <- predict(ogreg.3.add.2, newdata = pre.data3x, type = "response")
pre.data3x

  Strain Housing      prob
1     WT     STD 0.8040861
2     WT     ENR 0.8327355
3    dTg     STD 0.5768254
4    dTg     ENR 0.6231321
5 PS1dE9     STD 0.7611506
6 PS1dE9     ENR 0.7944739
7 APPswe     STD 0.6935843
8 APPswe     ENR 0.7330278

The full overview of the logistic regression results table is here.

Age group 17-25 acquisition Strain

ogistic.data.1725.acq=filter(ogistic.data.1725, `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity
simple model including housing only

ogreg.1725.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.1725.acq)

simple model including Strain only

ogreg.1725.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.1725.acq)

additive model including all factors

ogreg.1725.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.1725.acq)

interaction model including all factors

ogreg.1725.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.1725.acq)

Model comparisons using the anova function

anova(ogreg.1725.acq.add.1b, ogreg.1725.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       552     733.82                          
2       554     752.57 -2  -18.747 8.493e-05 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(ogreg.1725.acq.add.2, ogreg.1725.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df   Deviance Pr(>Chi)
1       551     733.82                       
2       552     733.82 -1 -0.0068548    0.934

adding factor housing does not improve the model

anova(ogreg.1725.acq.int, ogreg.1725.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       548     730.06                     
2       552     733.82 -4  -3.7677   0.4384

summary(ogreg.1725.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.1725.acq)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.4516  -1.3893   0.9260   0.9612   1.6394  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)  
(Intercept)              -0.1719     0.3393  -0.506   0.6125  
HousingSTD                0.6419     0.4431   1.448   0.1475  
StraindTg                -0.1959     0.4573  -0.428   0.6684  
StrainPS1dE9              0.6574     0.4649   1.414   0.1573  
StrainWT                  0.7967     0.3796   2.099   0.0358 *
HousingSTD:StraindTg     -1.3156     0.7183  -1.832   0.0670 .
HousingSTD:StrainPS1dE9  -0.6013     0.6479  -0.928   0.3534  
HousingSTD:StrainWT      -0.7342     0.5001  -1.468   0.1421  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 752.70  on 555  degrees of freedom
Residual deviance: 730.06  on 548  degrees of freedom
AIC: 746.06

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.1725.acq.int))

            (Intercept)              HousingSTD               StraindTg 
              0.8421053               1.9000000               0.8221154 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.9296875               2.2181604               0.2683179 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              0.5481159               0.4798810 

impact of predictor “Strain”

pre.data.17.acq <- data.frame(Strain=c("WT", "dTg", "PS1dE9", "APPswe"))
pre.data.17.acq$prob <- predict(ogreg.1725.acq.add.1b, newdata = pre.data.17.acq, type = "response")
pre.data.17.acq

  Strain      prob
1     WT 0.6400000
2    dTg 0.3582090
3 PS1dE9 0.6233766
4 APPswe 0.5517241

impact of predictor “Housing”

pre.data.17.acq <- data.frame(Housing=c("STD","ENR"))
pre.data.17.acq$prob <- predict(ogreg.1725.acq.add.1a, newdata = pre.data.17.acq, type = "response")
pre.data.17.acq

  Housing      prob
1     STD 0.5971731
2     ENR 0.5824176

Age group 17-25 reversal Strain

ogistic.data.1725.rev=filter(ogistic.data.1725, `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity
simple model including housing only

ogreg.1725.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.1725.rev)

simple model including Strain only

ogreg.1725.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.1725.rev)

additive model including all factors

ogreg.1725.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.1725.rev)

interaction model including all factors

ogreg.1725.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.1725.rev)

Model comparisons using the anova function

anova(ogreg.1725.rev.add.1b, ogreg.1725.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       376     319.30                          
2       378     357.43 -2  -38.128 5.254e-09 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model including Strain better than model including housing

anova(ogreg.1725.rev.add.2, ogreg.1725.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       375     316.76                     
2       376     319.30 -1  -2.5396    0.111

adding factor housing does not improve the model

anova(ogreg.1725.rev.int, ogreg.1725.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1       372      290.8                          
2       376      319.3 -4  -28.505 9.852e-06 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model

summary of minimal adequate model

summary(ogreg.1725.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.1725.rev)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-2.2035   0.4298   0.4895   0.4895   1.6176  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)
(Intercept)                17.57     989.05   0.018    0.986
HousingSTD                -16.18     989.05  -0.016    0.987
StraindTg                 -18.56     989.05  -0.019    0.985
StrainPS1dE9              -16.54     989.05  -0.017    0.987
StrainWT                  -15.23     989.05  -0.015    0.988
HousingSTD:StraindTg       19.07     989.05   0.019    0.985
HousingSTD:StrainPS1dE9    16.49     989.05   0.017    0.987
HousingSTD:StrainWT        15.91     989.05   0.016    0.987

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 363.07  on 379  degrees of freedom
Residual deviance: 290.80  on 372  degrees of freedom
AIC: 306.8

Number of Fisher Scoring iterations: 16

exp(coef(ogreg.1725.rev.int))

            (Intercept)              HousingSTD               StraindTg 
           4.254481e+07            9.401851e-08            8.705418e-09 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
           6.581296e-08            2.428812e-07            1.914517e+08 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
           1.443485e+07            8.087436e+06 

impact of predictor “Strain”

pre.data.17.revs <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.17.revs$prob <- predict(ogreg.1725.rev.int, newdata = pre.data.17.revs, type = "response")
pre.data.17.revs

  Strain Housing      prob
1     WT     STD 0.8870968
2     WT     ENR 0.9117647
3    dTg     STD 0.8695652
4    dTg     ENR 0.2702703
5 PS1dE9     STD 0.7916667
6 PS1dE9     ENR 0.7368421
7 APPswe     STD 0.8000000
8 APPswe     ENR 1.0000000

Age group 13-14 acquisition Strain x Housing

ogistic.data.acq.1314=filter(ogistic.data.1314,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.1314.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.acq.1314)

simple model including Strain only

ogreg.1314.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.acq.1314)

additive model including all factors

ogreg.1314.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.acq.1314)

interaction model including all factors

ogreg.1314.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.acq.1314)

comparisons between model using the anova function

anova(ogreg.1314.acq.add.1b, ogreg.1314.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       506     643.40                     
2       506     691.75  0  -48.349         

model with Strain only is good

anova(ogreg.1314.acq.add.2, ogreg.1314.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       505     641.78                     
2       506     643.40 -1  -1.6198   0.2031

adding factor housing also has good fit

anova(ogreg.1314.acq.int, ogreg.1314.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       504     638.28                       
2       506     643.40 -2  -5.1186  0.07736 .
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Strain model is best

summary of interactive model

summary(ogreg.1314.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.acq.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.7138  -1.0447   0.7235   0.9323   1.3654  

Coefficients:
                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)          -0.4318     0.1816  -2.377   0.0174 *  
HousingSTD            0.1113     0.2671   0.417   0.6769    
StrainWT              1.6386     0.2648   6.189 6.07e-10 ***
HousingSTD:StrainWT  -0.7099     0.3800  -1.868   0.0617 .  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 693.42  on 507  degrees of freedom
Residual deviance: 638.28  on 504  degrees of freedom
AIC: 646.28

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.1314.acq.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          0.6493506           1.1177419           5.1480000           0.4916989 

impact of predictors “Strain”

pre.data1314.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data1314.acq$prob <- predict(ogreg.1314.acq.add.1b, newdata = pre.data1314.acq, type = "response")
pre.data1314.acq

  Strain Housing      prob
1     WT     STD 0.7153285
2     WT     ENR 0.7153285
3    dTg     STD 0.4059829
4    dTg     ENR 0.4059829

impact of predictors “Housing”

pre.data1314.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data1314.acq$prob <- predict(ogreg.1314.acq.add.1a, newdata = pre.data1314.acq, type = "response")
pre.data1314.acq

  Strain Housing      prob
1     WT     STD 0.5414847
2     WT     ENR 0.5985663
3    dTg     STD 0.5414847
4    dTg     ENR 0.5985663

Age group 13-14 reversal Strain x Housing

ogistic.data.rev.1314=filter(ogistic.data.1314,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.1314.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.rev.1314)

simple model including Strain only

ogreg.1314.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.rev.1314)

additive model including all factors

ogreg.1314.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.rev.1314)

interaction model including all factors

ogreg.1314.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.rev.1314)

comparisons between model using the anova function

anova(ogreg.1314.rev.add.1b, ogreg.1314.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       360     337.23                     
2       360     384.85  0   -47.62         

anova(ogreg.1314.rev.add.2, ogreg.1314.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       359     337.11                     
2       360     337.23 -1 -0.11714   0.7322

adding factor does not improve model

anova(ogreg.1314.rev.int, ogreg.1314.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       358     333.32                     
2       360     337.23 -2  -3.9027   0.1421

Strain model is the best

summary of interaction model

summary(ogreg.1314.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.rev.1314)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-2.3200   0.3747   0.3747   0.9145   1.0913  

Coefficients:
                    Estimate Std. Error z value Pr(>|z|)    
(Intercept)           0.2059     0.2277   0.904   0.3659    
HousingSTD            0.4496     0.3288   1.367   0.1715    
StrainWT              2.4152     0.4312   5.601 2.13e-08 ***
HousingSTD:StrainWT  -1.1378     0.5884  -1.934   0.0532 .  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 384.89  on 361  degrees of freedom
Residual deviance: 333.32  on 358  degrees of freedom
AIC: 341.32

Number of Fisher Scoring iterations: 5

exp(coef(ogreg.1314.rev.int))

        (Intercept)          HousingSTD            StrainWT HousingSTD:StrainWT 
          1.2285714           1.5676141          11.1918605           0.3205376 

impact of predictors “Strain”

pre.data1314.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data1314.rev$prob <- predict(ogreg.1314.rev.add.1b, newdata = pre.data1314.rev, type = "response")
pre.data1314.rev

  Strain Housing      prob
1     WT     STD 0.9073171
2     WT     ENR 0.9073171
3    dTg     STD 0.6050955
4    dTg     ENR 0.6050955

impact of predictors “Housing”

pre.data1314.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data1314.rev$prob <- predict(ogreg.1314.rev.add.1a, newdata = pre.data1314.rev, type = "response")
pre.data1314.rev

  Strain Housing      prob
1     WT     STD 0.7710843
2     WT     ENR 0.7806122
3    dTg     STD 0.7710843
4    dTg     ENR 0.7806122

Age group 3 acquisition Strain x Housing

ogistic.data.acq.3=filter(ogistic.data.3,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.3.acq.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.acq.3)

simple model including Strain only

ogreg.3.acq.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.acq.3)

additive model including all factors

ogreg.3.acq.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.acq.3)

interaction model including all factors

ogreg.3.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.acq.3)

comparisons between model using the anova function

anova(ogreg.3.acq.add.1b, ogreg.3.acq.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance  Pr(>Chi)    
1      1398     1727.9                          
2      1400     1804.8 -2  -76.907 < 2.2e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model with Strain only is really good

anova(ogreg.3.acq.add.2, ogreg.3.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1      1397     1722.5                       
2      1398     1727.9 -1  -5.3748  0.02043 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

adding factor housing also has good fit

anova(ogreg.3.acq.int, ogreg.3.acq.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)   
1      1394     1712.5                        
2      1398     1727.9 -4  -15.461 0.003835 **
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

interaction model better than additive model interaction model is the best

summary of interaction model

summary(ogreg.3.acq.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.acq.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-1.7971  -1.2746   0.7551   0.9157   1.4284  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)               0.6523     0.1424   4.581 4.63e-06 ***
HousingSTD               -0.1133     0.2203  -0.514   0.6070    
StraindTg                -0.4268     0.2048  -2.084   0.0372 *  
StrainPS1dE9              0.4203     0.2147   1.958   0.0502 .  
StrainWT                  0.4566     0.2487   1.836   0.0663 .  
HousingSTD:StraindTg     -0.6852     0.3158  -2.170   0.0300 *  
HousingSTD:StrainPS1dE9  -0.1545     0.3173  -0.487   0.6264    
HousingSTD:StrainWT       0.3974     0.3519   1.129   0.2588    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 1807.0  on 1401  degrees of freedom
Residual deviance: 1712.5  on 1394  degrees of freedom
AIC: 1728.5

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.3.acq.int))

            (Intercept)              HousingSTD               StraindTg 
              1.9200000               0.8928571               0.6526104 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.5224359               1.5787760               0.5039771 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              0.8568804               1.4879242 

impact of predictors “Strain x Housing”

pre.data3.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data3.acq$prob <- predict(ogreg.3.acq.int, newdata = pre.data3.acq, type = "response")
pre.data3.acq

  Strain Housing      prob
1     WT     STD 0.8010753
2     WT     ENR 0.7519380
3    dTg     STD 0.3605442
4    dTg     ENR 0.5561497
5 PS1dE9     STD 0.6910112
6 PS1dE9     ENR 0.7450980
7 APPswe     STD 0.6315789
8 APPswe     ENR 0.6575342

Age group 3 reversal Strain x Housing

ogistic.data.rev.3=filter(ogistic.data.3,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

fitting different logistic regression models with increasing complexity simple model including housing only

ogreg.3.rev.add.1a <- glm(strategy.class ~ Housing, family = binomial, data = ogistic.data.rev.3)

simple model including Strain only

ogreg.3.rev.add.1b <- glm(strategy.class ~ Strain, family = binomial, data = ogistic.data.rev.3)

additive model including all factors

ogreg.3.rev.add.2 <- glm(strategy.class ~ Housing + Strain, family = binomial, data = ogistic.data.rev.3)

interaction model including all factors

ogreg.3.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = ogistic.data.rev.3)

comparisons between model using the anova function

anova(ogreg.3.rev.add.1b, ogreg.3.rev.add.1a, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Strain
Model 2: strategy.class ~ Housing
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)  
1       991     894.74                       
2       993     903.30 -2  -8.5607  0.01384 *
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

model with Strain only is really good (Pr=3.245e-13)

anova(ogreg.3.rev.add.2, ogreg.3.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing + Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       990     894.39                     
2       991     894.74 -1 -0.35067   0.5537

adding factor housing also has good fit (Pr=1.945e-5)

anova(ogreg.3.rev.int, ogreg.3.rev.add.1b, test="Chisq")

Analysis of Deviance Table

Model 1: strategy.class ~ Housing * Strain
Model 2: strategy.class ~ Strain
  Resid. Df Resid. Dev Df Deviance Pr(>Chi)
1       987     889.61                     
2       991     894.74 -4   -5.134   0.2738

Strain is best model

summary of interaction model

summary(ogreg.3.rev.int)

Call:
glm(formula = strategy.class ~ Housing * Strain, family = binomial, 
    data = ogistic.data.rev.3)

Deviance Residuals: 
    Min       1Q   Median       3Q      Max  
-2.0688   0.5003   0.5769   0.6242   0.7651  

Coefficients:
                        Estimate Std. Error z value Pr(>|z|)    
(Intercept)               1.7091     0.2371   7.207 5.73e-13 ***
HousingSTD               -0.4510     0.3222  -1.400   0.1615    
StraindTg                -0.6303     0.3093  -2.038   0.0416 *  
StrainPS1dE9              0.2681     0.3457   0.776   0.4380    
StrainWT                  0.3058     0.3882   0.788   0.4308    
HousingSTD:StraindTg      0.9091     0.4516   2.013   0.0441 *  
HousingSTD:StrainPS1dE9   0.1642     0.4789   0.343   0.7318    
HousingSTD:StrainWT       0.2081     0.5168   0.403   0.6872    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 903.56  on 994  degrees of freedom
Residual deviance: 889.61  on 987  degrees of freedom
AIC: 905.61

Number of Fisher Scoring iterations: 4

exp(coef(ogreg.3.rev.int))

            (Intercept)              HousingSTD               StraindTg 
              5.5238095               0.6369732               0.5324544 
           StrainPS1dE9                StrainWT    HousingSTD:StraindTg 
              1.3074713               1.3577586               2.4820511 
HousingSTD:StrainPS1dE9     HousingSTD:StrainWT 
              1.1783970               1.2313136 

impact of predictors “Strain”

pre.data3.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data3.rev$prob <- predict(ogreg.3.rev.add.1b, newdata = pre.data3.rev, type = "response")
pre.data3.rev

  Strain Housing      prob
1     WT     STD 0.8675799
2     WT     ENR 0.8675799
3    dTg     STD 0.7813765
4    dTg     ENR 0.7813765
5 PS1dE9     STD 0.8629630
6 PS1dE9     ENR 0.8629630
7 APPswe     STD 0.8146718
8 APPswe     ENR 0.8146718

impact of predictors “Housing”

pre.data3.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data3.rev$prob <- predict(ogreg.3.rev.add.1a, newdata = pre.data3.rev, type = "response")
pre.data3.rev

  Strain Housing      prob
1     WT     STD 0.8248945
2     WT     ENR 0.8368522
3    dTg     STD 0.8248945
4    dTg     ENR 0.8368522
5 PS1dE9     STD 0.8248945
6 PS1dE9     ENR 0.8368522
7 APPswe     STD 0.8248945
8 APPswe     ENR 0.8368522

2D - Age group 3

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
probs=read_excel("results_logistic_thresh.xlsx")
myColors = brewer.pal(4,"Set1")
myColors2= pal_npg("nrc")(3)
names(myColors)=levels(probs$Housing)
colScale=scale_colour_manual(name="Housing",values=myColors)
probs$`Experiment_Phase`=as.factor(probs$`Experiment_Phase`)
probs3=filter(probs, Age_group=='3')

gg = ggplot(probs3, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() + facet_grid(~Experiment_Phase) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 3 months old mice using Strategy class 1")
gg= gg+colScale


ggplotly(p=gg, height=400, width=800)

For ACQ Phase, Strain*Housing interaction model was used. For BOTH Phases, Strain+Housing additive model was used for predictions and for REV the simple strategy.class ~ Strain and strategy.class ~ Housing models were used separately.

2D - Age group 13-14

probs13=filter(probs, Age_group=='13-14')

gg = ggplot(probs13, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() + facet_grid(~Experiment_Phase) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 13-14 months old mice using Strategy class 1")
gg=gg+colScale
ggplotly(gg, height=400, width=800)

Strain*Housing interaction model was used for predictions of BOTH phases. In the ACQ and REV experiment phases, the simple models (strategy.class ~ Strain and strategy.class ~ Housing) showed the best fit. Take a look here

2D - Age group 17-25

probs17=filter(probs, Age_group=='17-25')

gg = ggplot(probs17, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_jitter() +scale_x_discrete(guide=guide_axis(n.dodge =2))+ facet_grid(~factor(Experiment_Phase, levels=c('ACQ','REV','BOTH'))) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 17-25 months old mice using Strategy class 1", caption="based on data from Govind & Meriem MWM Experiment in 2014")
gg=gg+colScale
gg

gg_bw=gg+theme_bw(base_size = 12)
gg_bw

gg_jco=gg_bw+scale_color_jco()
gg_jco

ggplotly(gg_jco, height=400, width=800)

For the ACQ AND BOTH phases, the simple strategy.class ~ Strain and strategy.class ~ Housing models were used and for the REV phase, the interactive Strain*Housing model showed the best fit and was therefore used for the predictions.

Strategy Legend

• 1 - thigmotaxis
  
• 2 - circling
  
• 3 - random path
  
• 4 - scanning
  
• 5 - chaining
  
• 6 - directed search
  
• 7 - corrected search
  
• 8 - direct path
  
• 9 - perseverance

Strategies 1-5 are less hippocampus-dependent, ‘unspatial’ and therefore get strategy.class = 0. In contrast to that, strategies 6-9 are more hippocampus-dependent, ‘spatial’ and get strategy.class = 1 assigned.

All probabilities in the following codes relate to choosing strategy.class = 1.

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
logistic.data.pub=read_excel("logistic_reg_thresh.xlsx")

Age group 17-25 Housing

logistic.data.pub.1725=filter(logistic.data.pub, Age_group1725=='1')

interaction model including all factors

logreg.pub.1725.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.1725)

interaction model

impact of predictor “Strain*Housing”

pre.data.pub.17 <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.pub.17$prob <- predict(logreg.1725.int, newdata = pre.data.pub.17, type = "response")
pre.data.pub.17

  Strain Housing      prob
1     WT     STD 0.5778894
2     WT     ENR 0.5915493
3    dTg     STD 0.3924051
4    dTg     ENR 0.2982456
5 PS1dE9     STD 0.5500000
6 PS1dE9     ENR 0.4891304
7 APPswe     STD 0.5641026
8 APPswe     ENR 0.4675325

Age group 13-14 Strain x Housing effects

logistic.data.pub.1314=filter(logistic.data.pub, Age_group1314=='1')

interaction model including all factors

logreg.1314.pub.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.1314)

impact of predictor “Strain*Housing”

pre.data.pub.1314 <- data.frame(Strain=c("WT", "WT","dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data.pub.1314$prob <- predict(logreg.1314.pub.int, newdata = pre.data.pub.1314, type = "response")
pre.data.pub.1314

  Strain Housing      prob
1     WT     STD 0.7416268
2     WT     ENR 0.8407407
3    dTg     STD 0.5215054
4    dTg     ENR 0.4536585

Age group 3 Strain x Housing

logistic.data.pub.3=filter(logistic.data.pub, Age_group3=='1')

interaction model including all factors

logreg.pub.3.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.3)

impact of predictors “Strain * Housing”

pre.data.pub.3x <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.pub.3x$prob <- predict(logreg.pub.3.int, newdata = pre.data.pub.3x, type = "response")
pre.data.pub.3x

  Strain Housing      prob
1     WT     STD 0.8217822
2     WT     ENR 0.8095238
3    dTg     STD 0.5615385
4    dTg     ENR 0.6355140
5 PS1dE9     STD 0.7533333
6 PS1dE9     ENR 0.8011364
7 APPswe     STD 0.6970803
8 APPswe     ENR 0.7303371

Age group 17-25 acquisition Strain

logistic.data.pub.1725.acq=filter(logistic.data.pub.1725, `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

interaction model including all factors

logreg.1725.pub.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.1725.acq)

impact of predictor “Strain*Housing”

pre.data.pub.17.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.pub.17.acq$prob <- predict(logreg.1725.pub.acq.int, newdata = pre.data.pub.17.acq, type = "response")
pre.data.pub.17.acq

  Strain Housing      prob
1     WT     STD 0.6300578
2     WT     ENR 0.6513158
3    dTg     STD 0.2608696
4    dTg     ENR 0.4090909
5 PS1dE9     STD 0.6285714
6 PS1dE9     ENR 0.6190476
7 APPswe     STD 0.6153846
8 APPswe     ENR 0.4571429

Age group 17-25 reversal Strain

logistic.data.pub.1725.rev=filter(logistic.data.pub.1725, `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

interaction model including all factors

logreg.pub.1725.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.1725.rev)

impact of predictor “Strain*Housing”

pre.data.17.pub.revs <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.17.pub.revs$prob <- predict(logreg.pub.1725.rev.int, newdata = pre.data.17.pub.revs, type = "response")
pre.data.17.pub.revs

  Strain Housing      prob
1     WT     STD 0.8870968
2     WT     ENR 0.9117647
3    dTg     STD 0.8695652
4    dTg     ENR 0.2702703
5 PS1dE9     STD 0.7916667
6 PS1dE9     ENR 0.7368421
7 APPswe     STD 0.8000000
8 APPswe     ENR 1.0000000

Age group 13-14 acquisition Strain x Housing

logistic.data.pub.acq.1314=filter(logistic.data.pub.1314,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

interaction model including all factors

logreg.pub.1314.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.acq.1314)

impact of predictors “Strain”

pre.data.pub.1314.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data.pub.1314.acq$prob <- predict(logreg.pub.1314.acq.int, newdata = pre.data.pub.1314.acq, type = "response")
pre.data.pub.1314.acq

  Strain Housing      prob
1     WT     STD 0.6475410
2     WT     ENR 0.7697368
3    dTg     STD 0.4205607
4    dTg     ENR 0.3937008

Age group 13-14 reversal Strain x Housing

logistic.data.pub.rev.1314=filter(logistic.data.pub.1314,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

interaction model including all factors

logreg.pub.1314.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.rev.1314)

impact of predictors “Strain”

pre.data.pub.1314.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg"),Housing=c("STD","ENR","STD","ENR"))
pre.data.pub.1314.rev$prob <- predict(logreg.pub.1314.rev.int, newdata = pre.data.pub.1314.rev, type = "response")
pre.data.pub.1314.rev

  Strain Housing      prob
1     WT     STD 0.8735632
2     WT     ENR 0.9322034
3    dTg     STD 0.6582278
4    dTg     ENR 0.5512821

Age group 3 acquisition Strain x Housing

logistic.data.pub.acq.3=filter(logistic.data.pub.3,  `_Day`=='1'|`_Day`=='2'|`_Day`=='3'|`_Day`=='4'|`_Day`=='5'|`_Day`=='6')

interaction model including all factors

logreg.pub.3.acq.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.acq.3)

impact of predictors “Strain x Housing”

pre.data.pub.3.acq <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.pub.3.acq$prob <- predict(logreg.pub.3.acq.int, newdata = pre.data.pub.3.acq, type = "response")
pre.data.pub.3.acq

  Strain Housing      prob
1     WT     STD 0.8010753
2     WT     ENR 0.7519380
3    dTg     STD 0.3605442
4    dTg     ENR 0.5561497
5 PS1dE9     STD 0.6910112
6 PS1dE9     ENR 0.7450980
7 APPswe     STD 0.6315789
8 APPswe     ENR 0.6575342

Age group 3 reversal Strain x Housing

logistic.data.pub.rev.3=filter(logistic.data.pub.3,  `_Day`=='7'|`_Day`=='8'|`_Day`=='9'|`_Day`=='10')

interaction model including all factors

logreg.pub.3.rev.int <- glm(strategy.class ~ Housing * Strain, family = binomial, data = logistic.data.pub.rev.3)

impact of predictors “Strain x Housing”

pre.data.pub.3.rev <- data.frame(Strain=c("WT","WT", "dTg","dTg", "PS1dE9","PS1dE9", "APPswe","APPswe"),Housing=c("STD","ENR","STD","ENR","STD","ENR","STD","ENR"))
pre.data.pub.3.rev$prob <- predict(logreg.pub.3.rev.int, newdata = pre.data.pub.3.rev, type = "response")
pre.data.pub.3.rev

  Strain Housing      prob
1     WT     STD 0.8547009
2     WT     ENR 0.8823529
3    dTg     STD 0.8230088
4    dTg     ENR 0.7462687
5 PS1dE9     STD 0.8442623
6 PS1dE9     ENR 0.8783784
7 APPswe     STD 0.7786885
8 APPswe     ENR 0.8467153

2D - Age group 3

setwd("C:/Google/Uni/Bachelorarbeit/R/Analysis")
probs.pub=read_excel("results_logistic_thresh_pub.xlsx")
myColors = brewer.pal(4,"Set1")
myColors2= pal_npg("nrc")(3)
names(myColors)=levels(probs.pub$Housing)
colScale=scale_colour_manual(name="Housing",values=myColors)
probs.pub$`Experiment_Phase`=as.factor(probs.pub$`Experiment_Phase`)
probs.pub3=filter(probs.pub, Age_group=='3')

gg = ggplot(probs.pub3, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_point() + facet_grid(~factor(Experiment_Phase, levels=c('ACQ','REV','BOTH'))) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 3 months old mice using Strategy class 1")
gg= gg+colScale
gg_bw=gg+theme_bw(base_size = 12)
gg_bw

ggplotly(p=gg, height=400, width=800)

2D - Age group 13-14

probs.pub13=filter(probs.pub, Age_group=='13-14')

gg = ggplot(probs.pub13, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_point() + facet_grid(~factor(Experiment_Phase, levels=c('ACQ','REV','BOTH'))) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 13-14 months old mice using Strategy class 1")
gg=gg+colScale
ggplotly(gg, height=400, width=800)

2D - Age group 17-25

probs.pub17=filter(probs.pub, Age_group=='17-25')

gg = ggplot(probs.pub17, mapping=aes(x=Strain, y=Probability, color=Housing)) + geom_point() +scale_x_discrete(guide=guide_axis(n.dodge =2))+ facet_grid(~factor(Experiment_Phase, levels=c('ACQ','REV','BOTH'))) + labs(x="Strain", y="P(strategy.class 1)", title="Probability of 17-25 months old mice using Strategy class 1", caption="based on data from Govind & Meriem MWM Experiment in 2014")
gg=gg+colScale
gg

gg_bw=gg+theme_bw(base_size = 12)
gg_bw

gg_jco=gg_bw+scale_color_jco()
gg_jco

ggplotly(gg_jco, height=400, width=800)

#Prediction Model explanation

Algorithm of strategy calling:

The Rtrack algorithm was exclusively trained with data from Morris water maze experiments with mice of different sex and genotypes, where multiple experts visually classified images of swim paths into different search strategies. Those 9 search strategies followed the scheme described at Rtrack Documentation - Strategy description and were classified at least twice until no discrepancies in the called strategies were left. Cases, in which the swim path could not be decided without doubt due to a mixture of strategies in one path or due to none of the 9 strategies being adequate, were discarded.

We then assigned each track the strategy class 0 or 1 depending on their hippocampus-dependency. The assignment was done following figure 3 in Garthe et. al. 2016, where ‘unspatial’ strategies were labled with strategy class 0 and ‘spatial’ strategies were labled with class 1.(Garthe et al., 2016)

Strategy.class

We´ve then done a logistic regression, which is a statistical model for a regression analysis. To do this, we first applied different logistic regression models with increasing complexity to our data and compared them using the ANOVA function. With this we choose the minimal adequate model, which is the simplest model, that describes the data appropiatly. With this model, we then could compute a prediction, which is the propability of a mouse of a specific experiment paradigm, choosing strategy class 1 and therefore using a hippocampus dependent strategy.

As the machine-learning algorithm of RTrack always chooses a strategy for a Trial, even if the algorithm is not quite sure, there are a lot of strategy calls, that the algorithm does not have a high confidence in. It therefore is of interest to only choose and only work with data, where a certain confidence threshold is given.

We have observed during testing that confidence scores above 0.4 are typically accurate and reproducible
RTrack - Documentation(Overall, 2020)

Following the recomendation of the RTrack documentation, a confidence threshold of 0.4 was choosen, which means that all strategies that did not were called with a confidence of 0.4 or higher, were discarded.(Overall, 2020)

13-14 Velocity Plot

mo14  %>%
mutate(mo14, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=mean.velocity, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2))+
  scale_linetype_manual(values=c(1,1,2,2))+
  stat_summary(fun.data = mean_se, geom = "errorbar")+
  labs(x="Day",
       y="Average Velocity [cm/s]",
       title="Mean Velocity 13-14mo mice")+theme_bw(base_size = 12)

## 3 mo Velocity Plot{#anchor54.15}

mo3  %>%
mutate(mo3, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=mean.velocity, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2,15,0,18,5))+
  scale_linetype_manual(values=c(1,1,2,2,3,3,4,4))+
  stat_summary(fun.data = mean_se, geom = "errorbar",size=.5)+
  labs(x="Day",
       y="Average Velocity [cm/s]",
       title="Mean Velocity 3 mo mice")+theme_bw(base_size = 12)

## Velocity Plots box{#anchor54.17}

Results.paper  %>%
ggplot(aes(x=`Age_group`,y=mean.velocity,fill=Age_group))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  scale_x_discrete(limits=c("3", "13-14"))+
  theme_classic(base_size = 12)+
  theme(legend.position = c(.5, .8))+
  labs(x="Age in months",
       y="Average Velocity [cm/s]",
       title="Mean Velocity all mice",
       fill="Age group")

## Latency Plots 13/14 {#anchor54.2}

mo14  %>%
mutate(mo14, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2,15,0,18,5))+
  scale_linetype_manual(values=c(1,1,2,2,3,3,4,4))+
  stat_summary(fun.data = mean_se, geom = "errorbar",size=.5)+
  theme_classic(base_size = 12)+
    geom_vline(xintercept =7, linetype="dotted", 
                color = "black", size=.5)+
  labs(x="Day",
       y="Latency [s]",
       title="Mean Latency to goal 13-14mo mice")

mo3  %>%
mutate(mo3, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=latency.to.goal, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2,15,0,18,5))+
  scale_linetype_manual(values=c(1,1,2,2,3,3,4,4))+
  stat_summary(fun.data = mean_se, geom = "errorbar",size=.5)+
  theme_classic(base_size = 12)+
  geom_vline(xintercept =7, linetype="dotted", 
                color = "black", size=.5)+
  labs(x="Day",
       y="Latency [s]",
       title="Mean Latency to goal 3mo mice")

Mean Path Length Plots 13

mo14  %>%
mutate(mo14, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2,15,0,18,5))+
  scale_linetype_manual(values=c(1,1,2,2,3,3,4,4))+
  stat_summary(fun.data = mean_se, geom = "errorbar",size=.5)+
  theme_classic(base_size = 12)+
    geom_vline(xintercept =7, linetype="dotted", 
                color = "black", size=.5)+
  labs(x="Day",
       y="Path length [cm]",
       title="Mean Path Length 13-14mo mice")

Mean Path Length Plots 3

mo3  %>%
mutate(mo3, `_Day`=fct_relevel(`_Day`,"1","2","3","4","5","6","7","8","9","10"))%>%
ggplot(aes(x=`_Day`,y=path.length, ,group=Condition))+
  geom_line(stat="summary", fun.y=mean,aes(linetype=Condition), size=1)+
  geom_point(stat="summary", fun.y=mean,aes(shape=Condition),size=2)+
  scale_colour_grey()+
  scale_shape_manual(values=c(16,1,17,2,15,0,18,5))+
  scale_linetype_manual(values=c(1,1,2,2,3,3,4,4))+
  stat_summary(fun.data = mean_se, geom = "errorbar",size=.5)+
  theme_classic(base_size = 12)+
  geom_vline(xintercept =7, linetype="dotted", 
                color = "black", size=.5)+
  labs(x="Day",
       y="Path length [cm]",
       title="Mean Path Length 3mo mice")

Probe Trial time in goal quadrant Plots

mo14.probe  %>%
ggplot(aes(x='',y=time.in.zone.n.quadrant, ,group=Condition,fill=Condition))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  labs(x="Condition",
       y="Time [s]",
       title="Probe Trial time in goal quadrant 13-14mo mice")

Probe Trial time in old goal zone Plots

mo14.probe  %>%
ggplot(aes(x='',y=time.in.zone.old.goal, ,group=Condition,fill=Condition))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  labs(x="Condition",
       y="Time [s]",
       title="Probe Trial time in old goal zone 13-14mo mice")

Probe Trial distance to old goal Plots

mo14.probe  %>%
ggplot(aes(x='',y=mean.d.old.goal, ,group=Condition,fill=Condition))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  labs(x="Condition",
       y="Distance [cm]",
       title="Probe Trial mean distance to old goal 13-14mo mice")

13mo Probe Trial old goal crossings Plots

mo14.probe  %>%
ggplot(aes(x='',y=old.goal.crossings, ,group=Condition,fill=Housing))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  theme(legend.position = c(.5, .7))+
  facet_grid(~Strain)+
  labs(x="Condition",
       y="Number of goal crossings",
       title="Probe Trial old goal crossings 13-14mo mice")

mo3.probe  %>%
ggplot(aes(x='',y=old.goal.crossings, ,group=Condition,fill=Housing))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  theme(legend.position = "none")+
  facet_grid(~Strain)+
  labs(x="Condition",
       y="Number of goal crossings",
       title="Probe Trial old goal crossings 3mo mice")

Results.paper.probe  %>%
ggplot(aes(x='Condition',y=old.goal.crossings, ,group=Condition,fill=Condition))+
  geom_boxplot()+
  scale_fill_brewer(palette="Greys")+
  theme_classic(base_size = 12)+
  facet_grid(~factor(Age_group))+
   labs(x="Condition",
       y="Goal crossings",
       title="Probe Trial old goal crossings all mice")

probs.pub.red=filter(probs.pub, Age_group=='3'|Age_group=='13-14')
probs.pub3=mutate(probs.pub3, `Strain`=fct_relevel(`Strain`,"WT","dTg","APPswe","PS1dE9"))
probs.pub13=mutate(probs.pub13, `Strain`=fct_relevel(`Strain`,"WT","dTg","APPswe","PS1dE9"))
probs.pub.red  %>%
mutate(probs.pub.red, `Strain`=fct_relevel(`Strain`,"WT","dTg","APPswe","PS1dE9"))%>%
ggplot()+
  geom_segment(data=probs.pub3, aes(x=Strain,xend=Strain, y=0.5,yend=Probability,color=Age_group))+
  geom_segment(data=probs.pub13, aes(x=Strain,xend=Strain, y=0.5,yend=Probability,color=Age_group))+
  geom_point(data=probs.pub.red, aes(x=Strain,y=Probability,color=Age_group,shape=Housing),size=2)+
  facet_grid(~factor(Experiment_Phase, levels=c('ACQ','REV','BOTH')))+
  scale_shape_manual(values=c(16,1))+
  
  coord_flip()+
  geom_hline(yintercept = .5, linetype="dotted", 
                color = "red", size=.5)

Different levels of Strain are associated with significant different path length.

P-value for the interaction between Housing*Strain is … –> relationship between Housing and Strain depends on the Housing method