Abstract
Mol Psychiatry. 2025 Nov 1. doi: 10.1038/s41380-025-03330-4. Online ahead of print.
ABSTRACT
Machine learning (ML) is revolutionising our ability to decode the complex genetic architectures of brain disorders. In this review we examine the strengths and limitations of ML methods, highlighting their applications in genetic prediction, patient stratification, and the modelling of genetic interactions. We explore how ML can augment polygenic risk scores (PRS) through advanced techniques and how integrating functional genomics and multimodal data can address challenges like rare variants and weak genetic effects. Additionally, we discuss the importance of embedding biological knowledge into ML models to enhance interpretability and uncover meaningful insights. With the ongoing expansion of phenotype-genotype datasets and advances in federated learning, ML is poised to compete with and surpass classical statistical methods in disease risk prediction and identifying genetically homogenous subgroups. By balancing the strengths and weaknesses of these approaches, we provide a roadmap for leveraging ML to unravel the genomic complexity of brain disorders and drive the next wave of discoveries.
PMID:41176580 | DOI:10.1038/s41380-025-03330-4
UK DRI Authors
