Connectivity-based Psychometric Prediction (CBPP)

Reference

Wu J, Eickhoff SB, Hoffstaedter F, Patil KR, Schwender H, Yeo BTT, Genon S. 2021. A connectivity-based psychometric prediction framework for brain-behavior relationship studies. Cerebral Cortex. 31(8): 3732-3751. https://doi.org/10.1093/cercor/bhab044.

Wu J, Eickhoff SB, Li J, Yeo BTT, Genon S. Replication (or not) of connectivity-based psychometric prediction patterns in distinct population neuroscience cohorts. In prep.

Background

The CBPP framework is an effort to summarize the general workflow of and systematically assess the common parameters in connectivity-based psychometric prediction studies. Our work consists of 2 aspects:

1. Whole-brain CBPP: In the preliminary analysis, we utilised all region-to-region connectivity values for prediction to find the overall best combination of approaches.

2. Region-wise CBPP: In order to improve the neurobiological validity (or interpretability) of psychometric prediction models, We propose a parcel-wise prediction approach, where models are trained on each parcel's connectivity profiles separately. We further illustrate 2 applications for this aproach:

   • single parcel's psychometric profile (below shows the psychometric profile of a pair of parcels in the Broca region)
   
   
   • single psychometric variable's predictive power variation across parcels (below shows the predictive power variation for 2-back task accuracy)
   
   

Later, focusing on the predictive power variation across parcels of fluid cognition and openness, we investigated the generalisability of such region-wise prediction patterns across distinct cohorts.

Replication

Please see the READMEs in the HCP_surface_CBPP and HCP_volume_CBPP folders on how to replicate the results in our 2021 paper.

Please see the README in the generalisability_CBPP folder on how to replicate the results in our 2022 paper.

Code Release

We release two Matlab functions, CBPP_wholebbrain.m and CBIG_parcelwise.m, for implementing any combination of approaches investigated in our paper.

Note that the connectivity and psychometric data should be prepared before using these functions. For computing preprocessing, connectivity, etc. as done in our paper, see the README in HCP_surface_CBPP folder. For help on extracting the psychometric data, see the README in bin/extraction_scripts folder.

To run whole-brain or parcel-wise CBPP, use the following commands in Matlab:

CBPP_wholebrain(fc, y, conf, cv_ind, out_dir, options)

CBPP_parcelwise(fc, y, conf, cv_ind, out_dir, options)

For more detailed usage for each function, use the following commands in Matlab:

help CBPP_wholebrain

help CBPP_parcelwise

Additional Information

1. Flowchart explaining the detailed workflow and cross-validation procedures are in the README file in the bin/procedure_descriptions folder.

2. Scripts used to compare different whole-brain CBPP approach and compute statistical significance for parcel-wise CBPP results can be found in the bin/evaluation_scripts folder. See the README in the folder for their usage.

Bugs and Questions

Please contact Jianxiao Wu at [email protected].