Editorial: Decoding SLE with machine learning: new tools to change the future of this rare condition

Systemic lupus erythematosus (SLE) represents an archetype of clinical and biological complexity. Its multi-organ involvement, relapsing–remitting course, and unpredictable response to therapy make the disease challenging both for patients and clinicians. Despite substantial progress in classification, therapeutic development, and treat-to-target strategies, major unmet needs persist, including earlier diagnosis, reliable prediction of disease course, identification of patients at risk of irreversible damage and recognition of different clusters for specific targeted therapies. SLE is not a disease that can be fully captured by single biomarkers, isolated organ domains, or linear associations. Rather, it entails interacting layers of genetic susceptibility, immune dysregulation, environmental triggers, metabolic rewiring, serologic activity, tissue injury and accumulating damage. In this context, machine learning (ML) provides a powerful framework to detect hidden patterns, integrate complex data layers, and move toward more precise disease stratification with possibly new therapeutic strategies.

This Research Topic was conceived to explore how ML may contribute to decoding SLE, from molecular discovery to clinical decision-making. The articles collected here illustrate complementary dimensions of this effort: patient phenotyping, biomarker discovery, immune-pathway characterization, multi-omics integration, and prediction of disease activity and treatment-related outcomes. These studies clearly show how ML may extend clinical inference, refine phenotyping, and generate new biologically meaningful hypotheses.

A first key message emerging from this Research Topic is that SLE heterogeneity can be reframed as a data-driven challenge. In this regard, Gatto et al. address a central issue in therapeutic stratification: patients selected for the same biologic therapy may in fact represent biologically and clinically distinct disease subgroups. In this study, unsupervised ML was applied to real-world data from the Lupus Italian REgistry (LIRE), focusing on patients initiating belimumab as add-on to standard therapy. Four distinct clusters emerged: renal/serologically active disease, neuropsychiatric-cardiovascular involvement with a higher damage burden, hematologic manifestations, and a lower-damage phenotype. Although clinical response to belimumab was broadly comparable across clusters, patients with renal/serologically active disease showed a relative higher response rate, whereas those with pre-existing damage remained at greater risk of chronic damage burden. This study demonstrates how unsupervised ML can reveal clinically meaningful heterogeneity within a treatment-defined population, shifting from a “one disease, one treatment” approach toward a stratified therapeutic reasoning.

A second key message is the potential use of ML to identify immune-related biomarkers. Zhang et al. used ML-assisted bioinformatics to identify immune-related diagnostic signatures in SLE. By integrating differential expression analysis, weighted gene co-expression network analysis, and multiple feature-selection algorithms, the authors identified a five-gene immune signature (CXCL2, CXCL8, FOS, NFKBIA, and CXCR2) associated with SLE. This signature was validated in external datasets and peripheral blood mononuclear cells, and was linked to immune-cell patterns, particularly neutrophil-related biology. The prominent role of chemokine signaling and neutrophil-related pathways highlighted in this work is noteworthy, given the growing recognition of innate immune activation, neutrophil extracellular traps, and interferon amplification in SLE pathogenesis. This study supports the potential of ML to identify specific immune signatures, improving our understanding of disease pathogenesis.

A third point is ML-assisted biomarker discovery. Ma et al. focused on CLEC4E, an innate immune receptor, as a potential novel biomarker in SLE. They integrated public transcriptomic data with clinical validation in a large case-control study and showed that CLEC4E expression was significantly increased in SLE patients compared with healthy controls. Moreover, it was strongly correlated with disease activity and with conventional immune markers, showing diagnostic and prognostic value, particularly when combined with established biomarkers such as complement levels, ANA, and anti-dsDNA antibodies. This study provides a clear example of how ML can drive biomarkers discovery and support the integration of novel biomarkers into clinical activity.

Finally, the study by Zhou et al. further broadens the landscape of ML application in SLE by adopting a multi-omics strategy, integrating transcriptomics, single-cell analysis, metabolomics and multiple ML methods. The authors sought to build an integrated disease model considering oxidative stress, metabolism, immune-cell dysfunction, and transcriptomic perturbation as connected layers of SLE pathogenesis. The identification of oxidative stress-related genes and metabolites provided a mechanistic framework in which immune dysfunction and metabolic remodeling emerged not as parallel observations but interconnected features of SLE biology. This study represents a robust example of how ML can serve as integrative tool to connect molecular pathways, immune-cell biology, and potential therapeutic targets.

Taken together, these studies show how ML is not a single method but a set of complementary analytical approaches. Unsupervised clustering can reveal patient subgroups; supervised models can identify diagnostic signatures; multi-omics pipelines can uncover mechanistic networks; and biomarkers-based models can improve disease diagnosis and activity assessment. Importantly, these approaches converge on a shared conclusion: SLE cannot be adequately described by one dimension alone. Clinical manifestations, organ damage, serological profile, immune-cell composition, gene expression, metabolism, and treatment exposure each provide only a partial view of the disease and should therefore be interpreted in an integrated model. ML becomes particularly valuable tool when used to connect these complementary dimensions.

However, this collection also emphasizes the need for caution. Many ML studies in SLE remain exploratory, retrospective, or based on limited cohorts. External validation, prospective and harmonized datasets, and clinically interpretable models are essential before ML can enter routine clinical care.

The value of ML will depend not only on accuracy metrics, but also on whether models are reproducible, explainable, equitable, and actionable for clinicians and patients. This Research Topic captures a field in transition. The promise of ML use in SLE is not simply to classify patients with greater precision, but to reveal latent architecture of the disease, the endotypes underlying clinical phenotypes, the molecular pathways beneath a heterogeneous syndrome. The future of ML in SLE should therefore move beyond prediction alone. Its greatest potential lies in supporting precision medicine, enabling earlier diagnosis, improved risk stratification, more rational selection of biologic therapies, and identification of novel therapeutic targets. Lupus nephritis, in which severe disease burden and incomplete response to therapy continue to drive morbidity, is a particularly important field for these developments.

In conclusion, the articles in this Research Topic provide a coherent view of how ML can help transform SLE research. They do not suggest that algorithms can replace clinical expertise, but rather they show how computational approaches can extend it. By integrating clinical and biological complexity, ML may help clinicians and researchers move from describing SLE heterogeneity to acting on it. This is the central promise of decoding SLE with machine learning, not simply to classify patients differently, but to change their future.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited and Reviewed by: John Harley, Independent Researcher, Cincinnati, United States