Reliable and generalizable brain-based predictions of cognitive functioning across common psychiatric illness

Reference

Chopra, S., Dhamala, E., Lawhead, C., Ricard, J., Orchard, E., An, L., Chen, P., Wulan, N., Kumar, P., Rubenstein. A., Moses, J., Chen, L., Levi, P., Aquino, K., Fornito, A., Harpaz-Rotem, I., Germine, L., Baker, J., Yeo, BT., Holmes, A. (2022) Reliable and generalizable brain-based predictions of cognitive functioning across common psychiatric illness. medRxiv.

---

Background

A primary aim of computational psychiatry is the establishment of predictive models linking individual differences in brain functioning with clinical symptoms. In particular, cognitive impairments are transdiagnostic, treatment resistant, and contribute to poor clinical outcomes. Recent work suggests thousands of participants may be necessary for the accurate and reliable prediction of cognition, calling into question the utility of most patient collection efforts. Here, using a transfer-learning framework, we train a model on functional imaging data from the UK Biobank (n=36,848) to predict cognitive functioning in three transdiagnostic patient samples (n=101-224). The model generalizes across datasets, and network-level brain features driving predictions are consistent between populations, with decreased functional connectivity within transmodal cortex and increased connectivity between unimodal and transmodal regions reflecting a transdiagnostic predictor of cognition. This work establishes that predictive models derived in large population-level datasets can be exploited to boost the prediction of cognitive function across clinical collection efforts.

Code and Data release

Code

The scripts folder contains the two folders: analysis and visualisation:

• The analysis folder contains three sub-folders and a conda environment file used for all analyses (predictingCognition_env.yml):
  • accuracy - contains python scripts that execute both meta-matching (compute_MM_cognitionPC.py and kernel ridge regression models (compute_KRR_cognitionPC.py). The master_run.sh script executes all python scripts with flags to indicating study sample and covariate regression. The nulls_runHPC folder contains scripts used to generate null prediction models and were executed on a High Performance Computing (HPC) cluster.
  • genralizibility - contains two python scripts used to test the genralizibility of the meta-matching (mm_cogs_generalize.py) and kernel ridge regression (krr_cogs_generalize.py) models (i.e. train on one full dataset and test on another independent dataset).
  • feature weights - contains a python script to generate spatially auto-correlated nulls (spin test), and conduct significance testing on the feature weights generated.
• The visualisation folder contains a Rmarkdown file (figures.Rmd) used to generate each main test and supplementary figure included in the paper. Each code chunk corresponds to a figure or panel. The brain renderings in Fig4 require a python env with both pyvista and pysurfer working smoothly (good luck!) Some examples below:

---

Data

• The primary data used are brain (419 x 419) FC matrices and cognitive functioning scores for three data sets:

  • Human Connectome Project - Early Psychosis (HCP-EP; n=145)

  • Transdiagnostic Connectomes Project (TCP; n=101)

  • Consortium for Neuropsychiatric Phenomics (CNP; n=224)

• FC matrices and cognitive functioning principal component (PC) scores for TCP and CNP can likely soon be shared openly (pending publication of the TCP dataset release paper). These files are >100mb and once I have approval I will uploaded here, but please reach out if you would like these, and I will send it through. The HCP-EP data requires data access permission and if you have this, we are happy to share the data used here.

• All model outputs reported in the paper and used to generate all figures are provided in output folder.

---

Meta-matching model

If you want to apply the meta-matching model to your own data, please see: https://github.com/ThomasYeoLab/Meta_matching_models

You will need functional coupling/connectivity (FC) data and a phenotype you want to predict. The FC data will need to be extracted using the 419 region atlas defined in the link above (also see data/atlas folder for a template).

Note: The meta-matching model use in the current analysis was run using data with GSR and z-scoring, so a version of the V1.1 meta-matching model was used.

---

Questions

Please contact me (Sidhant Chopra) at sidhant.chopra@yale.edu and/or sidhant.chopra4@gmail.com