A machine vision based frailty index for mice

Table of Contents

• Features
• Requirements
• Load packages
• Load functions
• Load data
• Preprocessing
• Run a synthetic experiment
• Sample code
• Train models
• Plot results
• References

Features

The scripts to generate features used here originate as output from this other repo: https://github.com/KumarLabJax/vFI-features.

Requirements

This program requires the following R packages:

reshape
lme4
caret
RLRSim
grf
rfinterval
mgcv
quantreg
ALEPlot
iml
ggplot
ggrepel

Load packages

packages <- c("reshape","lme4","caret","RLRSim","grf","rfinterval","mgcv","quantreg","ALEPlot","iml","ggplot","ggrepel")
lapply(packages, require, character.only = TRUE)

Load functions

source("vFI_expts.R")
source("vFI_functions.R")
source("vFI_utils.R")

Load data

data <- read.csv("Data/data.csv", header = TRUE, stringsAsFactors = FALSE)
data_frailty <- read.csv("Data/data_frailty_parameters.csv", header = TRUE, stringsAsFactors = TRUE)

Preprocessing

df <- read_data(data = data, summary_phenotypes = "median", type = "")
df <- preprocess_data(df, type = "")$df

df_frailty <- read_data(data = data_frailty, type = "frailty")
df_frailty <- preprocess_data(df_frailty, type = "frailty")

Run a synthetic experiment

#Experiment I: Test the performance of the video features in predicting FI and TestAge

results <- synthetic_experiment(nsim = 50, expt = "I", data = df)
save_results(results, expt = "I")

Sample code

Please refer to vFI.R as the starting point for running all synthetic experiments, which we recommend running on a high-performance computing cluster.

Train models

Please refer to vFI_functions.R and vFI_expts.R to see how the models are trained. yname is the dependent variable you're interested in predicting.

models <- fit_models(traindata, yname, expt = "I")
results <- performance_measure(models, testdata, yname, expt = "I")
save_results(results, expt = "I") #saves mae, rmse and r2 measures as csv files. These csv files are used as inputs to the functions below

plot_results(mae, rmse, r2, yname, expt = "I") 
plot_pred_int(mae, yname, type = "") #plot prediction intervals
interpretable_ML_plot(yname, mae, type) #type = c("feature_importance", "marginal", interaction")

Plot results

Please refer to vFI_Figures.R to recreate Figures 4,5 and Supplemental Figure 1 in the main manuscript.

References

Please refer to the main manuscript for all relevant references.