Data-driven dissection of heterogeneity: a computational framework for identifying depression subtypes through multi-omics and explainable AI

Modern psychiatry is shifting from unitary diagnostic models toward identifying biologically distinct depression subtypes. Despite the potential of multi-omics and AI, the field is hindered by non-standardized pipelines and poor reproducibility. We propose a four-pillar computational framework to standardize the subtyping process: (1) standardized preprocessing and feature embedding to ensure data integrity; (2) integrative multi-omics modeling strategies tailored to diverse sample sizes; (3) robust subtype identification and Explainable Artificial Intelligence (XAI) interpretation, where we propose the Minimum Reporting Standards for Computational Psychiatry Subtyping (MiR-CPS) to ensure methodological transparency; and (4) hierarchical clinical validation to benchmark subtype stability and utility. Beyond this core trajectory, we extend the framework to longitudinal trajectories and cross-diagnostic approaches to address temporal and diagnostic heterogeneity. This framework provides a reproducible roadmap for transitioning from raw high-dimensional data to clinically actionable subtypes, advancing evidence-based precision psychiatry.