Functional brain networks predicting different symptoms of schizophrenia based on connectome-based predictive...

Previous findings on brain functional alterations across different symptoms of schizophrenia (SCZ) patients had yielded inconsistent results. Small sample sizes could contribute to this inconsistency. To overcome this limitation, we conducted a multi-site study to explore the neural mechanisms underlying different symptoms in SCZ. This multi-site study included four datasets from three sites, comprising 258 SCZ patients and 222 healthy controls. A four-factor model based on the Positive and Negative Syndrome Scale (PANSS) was applied to identify four symptom dimensions of SCZ: negative, positive, emotional, and cognitive symptoms. Connectome-based predictive modeling (CPM) and node-based network analysis were conducted. Then, the support vector machine was used to classify SCZ patients and HCs. CPM models could successfully predict negative, positive, affective, and cognitive symptoms in SCZ patients, with correlation coefficients ranging from -0.339 to -0.057. Models for negative and affective symptom prediction were validated by two independent SCZ cohorts. Most predictive edges were connected between the Motor/Sensory (Mot), Fronto-Parietal, Default Mode, Salience, and other networks. The Mot network was involved in the CPM models across all symptom dimensions. Of the predictive edges, three edges exhibited increased FC, while six ones demonstrated decreased FC compared to HCs. These abnormal FCs could classify patients and HCs with an accuracy of 91.2 %. The predictive networks were primarily involved in sensory processing and high-level cognition, which could be a functional basis of SCZ. The Mot network may serve as a key hub across all symptom dimensions.