Addressing bias in biomarker discovery for inflammatory bowel diseases: A multi-faceted analytical approach

Xiang-Guang et al. investigate the identification of novel biomarkers linked to M1 macrophage infiltration in inflammatory bowel diseases (IBD). Utilizing advanced bioinformatics and machine learning techniques, the researchers developed predictive models and employed the SHAP algorithm to assess feature importance, revealing that the top ten features corresponded exclusively to host genes. However, significant concerns regarding the model-specific nature of SHAP assessments raise doubts about the reliability of feature importance. To address these issues, we advocate for a multifaceted approach combining feature agglomeration (FA), highly variable gene selection (HVGS), and Spearman's correlation for a more accurate analysis. This integrated methodology aims to enhance our understanding of biological factors in IBD and improve diagnostic and therapeutic strategies.