Major depressive disorder (MDD) is the most prevalent mental disorder that constitutes a major public health problem. A tool for predicting the risk of MDD could assist with the early identification of MDD patients and targeted interventions to reduce the risk. We aimed to derive a risk prediction tool that can categorize the risk of MDD as well as discover biologically meaningful genetic variants. Data analyzed were from the fourth and fifth data collections of a longitudinal community-based cohort from Southwest Montreal, Canada, between 2015 and 2018. To account for high dimensional features, we adopted a latent topic model approach to infer a set of topical distributions over those studied predictors that characterize the underlying meta-phenotypes of the MDD cohort. MDD probability derived from 30 MDD meta-phenotypes demonstrated superior prediction accuracy to differentiate MDD cases and controls. Six latent MDD meta-phenotypes we inferred via a latent topic model were highly interpretable. We then explored potential genetic variants that were statistically associated with these MDD meta-phenotypes. The genetic heritability of MDD meta-phenotypes was 0.126 (SE = 0.316), compared to 0.000001 (SE = 0.297) for MDD diagnosis defined by the structured interviews. We discovered a list of significant MDD - related genes and pathways that were missed by MDD diagnosis. Our risk prediction model confers not only accurate MDD risk categorization but also meaningful associations with genetic predispositions that are linked to MDD subtypes. Our findings shed light on future research focusing on these identified genes and pathways for MDD subtypes.
© 2022. The Author(s).