ImageNomer: developing an fMRI and omics visualization tool to detect racial bias in functional connectivity

Orlichenko, Anton; Daly, Grant; Wang, Yu-Ping; Liu, Anqi; Shen, Hui; Deng, Hong-Wen; Zhou, Ziyu

doi:10.36227/techrxiv.21992006.v1

ImageNomer_RaceBiasPaper.pdf (2.65 MB)

ImageNomer: developing an fMRI and omics visualization tool to detect racial bias in functional connectivity

preprint

posted on 2023-02-07, 21:43 authored by Anton OrlichenkoAnton Orlichenko, Grant Daly, Yu-Ping Wang, Anqi Liu, Hui Shen, Hong-Wen Deng, Ziyu Zhou

We use the Philadelphia Neurodevelopmental Cohort (PNC) dataset to identify that intelligence prediction using fMRI data is almost entirely dependent on racial confounds. Race prediction using fMRI connectivity data (85%) is more effective that sex prediction (78%), while intelligence prediction using within-race groups reveals no advantage over the null model. This is surprising because race is not a feature that has traditionally been predicted using connectivity data. The PNC dataset is available to research groups on request from the database of genotypes of phenotypes under ascession ID phs000607.v3.p2 Neurodevelopmental Genomics: Trajectories of Complex Phenotypes. Linear models (Ridge or Logistic Regression) were used throughout on correlation-based connectivity data and SNPs. All required aprovals were obtained.