Machine learning in graft-versus-host disease: current applications and future frontiers

Allogeneic hematopoietic stem cell transplantation (allo-HSCT) may be curative; however, after which graft-versus-host disease (GVHD) remains a leading cause of mortality. For the sake of high heterogeneity and the absence of widely used clinical biomarkers, clinicians are faced with obstacles when diagnosing and prognosticating GVHD. Current diagnostic approaches primarily rely on typical clinical manifestations, supplemented by laboratory indices such as inflammatory cytokine levels, often resulting in suboptimal efficacy. The emergence of machine learning (ML), however, offers a promising alternative, leveraging its unique advantages and powerful algorithms beyond traditional statistical tools. Thus, it is able to integrate existing findings on GVHD to build models for biomarker identification, diagnosis, and even prediction. In this review, we conclude present applications of ML for GVHD, exhibiting its distinctive qualitative and quantitive analytical capabilities in mining associated factors and bridging them with GVHD clinical outcomes. Additionally, we discuss future directions in this interdisciplinary field, combining the latest GVHD research frontiers with advanced bioinformatics methods, with the aim of providing guidance and insights for future progress.