Fig. 9
Specific brain states are predictive of distinct clinical symptoms in novel individuals. a State B4 uniquely predicted the presence of active psychosis in patients with psychotic illness. Results from a leave-one-subject-out cross-validation elastic net logistic regression analysis comparing predicted and observed psychotic symptoms (n = 91). The displayed area under the receiver-operating characteristic curve (AUC) reflects the scalar probability measure for each individual, predicting the likelihood of active psychotic symptoms. The blue semitransparent lines reflect 1000 ROC curves from 1000 permutation experiments. The black dashed line indicates the mean of these 1000 curves, approximating a null curve. The purple, red, green, and orange curves correspond to states A4, B4, C4, and D4, respectively. b Connectivity models defined on training data predict the presence of active psychotic symptoms in an independent group of participants (n = 31). The displayed AUC graph reflects the scalar probability measure, predicting the likelihood of active psychosis, defined using positive edges from the initial training set without further fitting or modification, in a held-out sample of patients
