Eur Arch Psychiatry Clin Neurosci. 2024 Sep 23. doi: 10.1007/s00406-024-01878-w. Online ahead of print.

ABSTRACT

Bipolar disorder (BPD) affects approximately 2% of the global population. Its clinical course is highly variable and current treatments are not always effective for all patients. Genetic factors play a significant role in BPD and its treatment, although the genetic background appear to be highly heterogeneous. Polygenic risk scores (PRS) are a powerful tool for risk assessment, yet using all genomic data may introduce confounding factors. Focusing on specific genetic clusters PRS (gcPRS) may mitigate this issue. This study aims to assess a neural network model’s efficacy in predicting response to treatment (RtT) in BPD individuals using PRS calculated from specific gcPRS and other variables. 1538 individuals from STEP-BD (age 41.39 ± 12.66, 59.17% female) were analyzed. gcPRS were calculated from a Genome-wide association study (GWAS) with clinical covariates and a molecular pathway analysis (MPA) based on drugs interaction networks. A neural network was trained using gcPRS and clinical variables to predict RtT. Ten biological networks were identified through MPA, with gcPRS derived from risk variants within corresponding gene groups. However, the model did not show significant accuracy in predicting RtT in BPD individuals. RtT in BPD is influenced by multiple factors. This study attempted a comprehensive approach integrating clinical and biological data to predict RtT. However, the model did not achieve significant accuracy, possibly due to limitations such as sample size, disorder complexity, and population heterogeneity. This data highlights the challenge of developing personalized treatments for BPD and the necessity for further research in this area.

PMID:39313733 | DOI:10.1007/s00406-024-01878-w