Psychopathology Prediction

A program appling a non-negative matrix factorization (NMF)-based machine learning framework to detect the overlapping communities in the brains of individuals with bipolar disorder. The abstract have been received by The Organization for Human Brain Mapping (OHBM), the annual conference.

Resting-state fMRI scans from 108 bipolar disorder patients retrieved from the Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP) database were employed. Psychopathology assessed by the Young Mania Rating Scale was then compressed into three dimensions (Vigor, Aggression, and Psychotic) using a NMF-based algorithm. Time-series of the fMRI data were extracted based on a parcellation system after combining Schäfer’s 400 cortical parcels with the 36 brainnetome subcortical parcels. Parcel-wise functional connectome was constructed for each individual by using a non-negative adaptive sparse representation approach. The symmetric non-negative matrix factorization was applied to individual connectomes to detect overlapping communities. Individual loadings on these communities were afterward fed into a regression vector model with stratified 10-fold cross-validation to predict individual scores of the three psychopathological dimensions. 

Fig 2. final poster on OHBM with further analysis