Angle Basis: a Generative Model and Decomposition for Functional Connectivity
In this paper we present a new decomposition and generative model for functional connectivity (FC), achieve 10x data compression, 97.3% identifiability, and modest improvent on FC-based predictive tasks. Additionally we are able to generate synthetic subjects with user-inputted clinical characteristics.
fMRI and phenotype data came from the Neurodevelopmental Genomics: Trajectories of Complex Phenotypes database of genotypes and phenotypes repository, dbGaP Study Accession ID phs000607.v3.p2, as well as the Bipolar and Schizophrenia Network for Intermediate Phenotypes study (http://b-snip.org/).
Additional data from OpenNeuro study ds004144 on fibromyalgia is included in the linked-to code.
The authors would like acknowledge the NIH (grants R01 GM109068, R01 MH104680, R01MH107354, P20 GM103472, R01 REB020407, R01 EB006841, R56 MH124925) and NSF(#1539067) for partial funding support
Email Address of Submitting Authoraorlichenko@tulane.edu
ORCID of Submitting Author0000-0001-7870-4970
Submitting Author's InstitutionTulane University
Submitting Author's Country
- United States of America