MACHINE LEARNING IN PSYCHIATRIC GENOMICS: INTEGRATING MULTIPLE LEVELS OF ANALYSIS THROUGH DEEP NETWORK MODELS

Abstract

Schizophrenia patients (SCZ) display widespread cognitive deficits that are strongly associated with functional outcomes. Cognitive impairments occur along a genetic continuum among SCZ, their unaffected first-degree relatives (FRs) and healthy controls (HCs). Although SCZ impairs the premorbid intelligence quotient (IQ) and causes a subsequent intelligence decline (ID), a decrease in present IQ from the premorbid level, it remains unclear when during the illness course these impairments develop. Differences in premorbid and present IQ and ID were investigated among 125 SCZ, 61 FRs and 107 HCs, using analysis of covariance and a paired t-test. Furthermore, these subjects were classified into preserved and deteriorated IQ groups based on the degree of ID, and we investigated which factors contribute to this classification. We found significant differences in premorbid and present IQ among the diagnostic groups. Compared with HCs, SCZ and FRs displayed lower premorbid and present IQ. There was no significant difference in premorbid IQ between SCZ and FRs, but SCZ had a significantly lower present IQ than FRs. Only SCZ showed a significant ID. As most FRs and HCs did not display an ID, there were fewer subjects with deteriorated IQ among FRs and HCs than among SCZ. Subjects with preserved IQ showed higher educational attainment than those with deteriorated IQ among SCZ and FRs. These findings suggest that the impairment of premorbid IQ and the ID in SCZ become evident before and around the time of onset, respectively, and different pathophysiological mechanisms might be related to these impairments.