Key points are not available for this paper at this time.
Systemic lupus erythematosus (SLE) is a clinically and biologically heterogeneous autoimmune disease in which patients with similar or disparate clinical phenotypes can exhibit distinct molecular drivers. This heterogeneity limits the utility of traditional, largely linear, analytic approaches and contributes to variations in diagnosis, prognosis, and therapeutic outcome. Artificial intelligence (AI), including machine learning (ML) and deep learning (DL), offers a complementary framework for extracting nonlinear patterns from complex biomedical datasets and for translating high dimensional molecular measurements into actionable clinical signatures. Here, we review how supervised and unsupervised ML methods are being applied to SLE, with a focus on multi-omics integration across genomics, transcriptomics, proteomics, and metabolomics. In this review, we specifically focus on how these approaches can define molecular endotypes, explaining heterogeneity in disease course and therapeutic response. We summarize evidence that AI-driven endotyping can reproducibly separate patients into molecularly defined subgroups dominated by distinct immune signatures and facilitate biomarker discovery and improved risk modelling. We highlight the limitations of conventional clinical phenotyping and the need for AI-driven biologically grounded patient stratification. Importantly, we frame these advances within a systems biology perspective, in which AI-driven integration of multi-omics and clinical data enables a unified approach to disease diagnosis, molecular stratification, prognosis, and therapeutic response prediction in SLE. Despite these advances, most AI-derived predictors remain research-grade, and require prospective multi-center validation and explicit demonstration of clinical utility before routine deployment. If key barriers to clinical translation can be overcome, AI-based methods hold promise for achieving truly personalized approaches in the management of SLE.
Biswas et al. (Thu,) studied this question.