TIGP (BIO)—Polygenic Prediction of Antidepressant Treatment Response

Abstract

Major Depressive Disorder (MDD) presents a significant clinical challenge, with only one-third of patients achieving remission following first-line antidepressant treatment. While antidepressant response and resistance exhibit substantial heritability—estimated at approximately 60%—identifying reliable genetic predictors has remained elusive. Early candidate gene studies, such as those focusing on 5-HTTLPR, have largely been dismissed as false positives by large-scale meta-analyses. Recent genome-wide association studies (GWAS) confirm that antidepressant response is a highly polygenic trait influenced by thousands of variants with minute effect sizes, requiring innovative computational approaches beyond single-locus associations.
We utilized a multi-polygenic score (multi-PGS) framework to enhance predictive power by leveraging the joint signal from multiple discovery GWASs. To address the historical lack of diverse genomic data, we employed PRS-CSx, a cross-population polygenic prediction method that integrates GWAS summary statistics from East Asian and European ancestries. We evaluated this approach across two Taiwanese cohorts: patients receiving selective serotonin reuptake inhibitors (SSRIs) and patients with treatment-resistant depression (TRD) receiving low-dose ketamine infusion. Models were built using 108 distinct polygenic scores representing psychiatric, physiological, and inflammatory traits.
In the SSRI cohort, the integration of multi-PGS into a baseline clinical model increased the adjusted R^2 from 0.097 to 0.172 for predicting symptom improvement at week 4. Significant genetic predictors included insomnia, multisite chronic pain, and autonomic nervous system regulation. For ketamine infusion treatment, the multi-PGS model improved the R^2 from 0.019 to 0.149, achieving an AUC of 0.7 for responder classification. Notably, inflammatory biomarkers showed a positive correlation with ketamine response, a direction opposite to that observed with SSRIs.
Our findings demonstrate the potential of multi-PGS and machine learning to bridge the gap between complex genetic architectures and clinical utility in precision psychiatry. Future directions include incorporating clinical and multi-omics data to further refine treatment response prediction.

Download