Model training/testing code, analyses, and visualizations for a project examining the relationships between BrainAGE and maturational metrics.
This repository contains code for the training and testing of a novel BrainAGE adolescent BrainAGE model, as well as code for the analyses and visualizations found in Whitmore, Weston, & Mills (https://doi.org/10.1162/imag_a_00037)
If you're interested in using the model to make predictions on your own data, follow the instructions below:
-
Download the model file: UpdatedModels/fit_workflow_combat.rds for baseline model, or UpdatedModels/fit_workflow_combat_followup.rds for follow-up model.
-
Prepare your own data. To make predictions, you should have a file with a column for age, and then columns for the brain features listed in feature_list.txt. Make sure your column names match up with the feature list. If you need to convert the names of your data columns, there are a few examples contained in the files within this repository. For converting from ABCD output, use the code in DataSetup.Rmd [lines 73-287]. For converting from Freesurfer output, use FreeSurferSetupBrainAGE.Rmd as a template (in progress, may be a few remaining bugs).
-
Once you've confirmed your data is in the correct format, run ABCDModel.Rmd (for baseline) or ABCDModelModelTraining_Followup.Rmd (lines 221-304, for follow-up) to get new predictions.
-
To perform bias correction, run lines 130-142 of ABCDModel.Rmd (baseline) and lines 253-288 of ABCDModelModelTraining_Followup.Rmd (follow-up).
If you want to train your own model, you can modify or use the code from UpdatedModels/ABCDModelTraining.Rmd (h/t to Vlad Drobinin for the original model training code (linked in /DevelopmentalBrainAGE), which has been updated to use workflows).
Additionally, some are some potential issues with predicting from the Drobinin et al., 2022 model used in the paper due to package deprecation/version issues. If you're interested in using that model, check out the DrobininModelPredictions.Rmd file, which contains some potentially helpful workarounds and instructions for predicting with that model!
Finally, if you have any questions about the code or project, or encounter any bugs, please reach out to me at lwhitmor@uoregon.edu