Blood-based gene signatures associated with therapeutic response to anti-TNF therapy in rheumatoid arthritis: a combined meta-analytical and machine learning approach.

OBJECTIVES

Tumour necrosis factor alpha (TNF-α) inhibitors have significantly improved outcomes in rheumatoid arthritis (RA); however, up to 30% to 40% of patients show inadequate response. This highlights the need for predictive biomarkers to guide personalised therapy. Blood transcriptome profiling is a promising strategy for identifying gene signatures linked to treatment response, though prior studies have shown inconsistent results due to technical and biological heterogeneity.

METHODS

We performed a comprehensive meta-analysis of baseline blood transcriptome datasets from 8 independent RA cohorts to identify genes consistently upregulated in patients who responded to TNF-α inhibitors. This analysis identified a core set of 39 recurrent genes (the recurrent gene set). To evaluate its predictive relevance, we trained machine learning models using 4 gene subsets: the recurrent set, the patent gene set derived from a biomarker patent, the top-ranked gene set based on Cohen's d effect size, and the full gene set. External validation was performed using an independent dataset.

RESULTS

The recurrent gene set showed strong and consistent predictive performance across machine learning models. Notably, several genes overlapping with the patent and top-ranked sets were among the top contributors, supporting their potential value for future diagnostic applications.

CONCLUSIONS

Our study demonstrates the feasibility of transcriptome-guided prediction of response to TNF-α inhibitors in RA. The integration of meta-analysis and machine learning provides a strong foundation for the development of precision diagnostic tools to support personalised treatment strategies in clinical practice.