From Statistical Correlation to Explainable AI: Reinterpreting Microbiome Biomarkers in Chronic Diseases Using SHAP

From Statistical Correlation to Explainable AI: Reinterpreting Microbiome Biomarkers in Chronic Diseases Using SHAP

Afagh Bapirzadeh1 Samaneh Rostami Safarloo2 Yasaman Rostami Safarloo3 Shadi Shahsavari4 Mohaddeseh Sabzikar Ghane5 Hasti Mostafavi Ghamsari6 Zarrin Minuchehr27

1) Ph.D. Candidate in Molecular Genetics, Department of Biology, Tehran North Branch, Islamic Azad University, Tehran, Iran Email:
2) B.Sc. Student in Microbiology, Department of Microbiology, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran Email:
3) B.Sc. Student in Microbiology, Department of Microbiology, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran Email:
4) B.Sc. Student in Microbiology, Department of Microbiology, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran Email:
5) B.Sc. Student in Microbiology, Department of Microbiology, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran Email:
6) B.Sc. Student in Microbiology, Department of Microbiology, Shahr-e-Qods Branch, Islamic Azad University, Tehran, Iran Email:
7) Associate Professor of Biophysics, National Institute of Genetic Engineering and Biotechnology (NIGEB), Tehran, Iranو

IOI Code : AAHD-ACZCAC
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    By using SHAP (SHapley Additive exPlanations), one of the most widely used approaches to explainable artificial intelligence (XAI), it becomes possible to identify and interpret the contribution of individual variables to machine-learning model predictions. This approach can mitigate the limitations of black-box models in biological research. In microbiome studies, SHAP can facilitate a more comprehensive understanding of how microbial features contribute to disease onset, progression, and prediction. Alterations in the composition and functional capacity of the human microbiome have been associated with the pathogenesis of several chronic diseases, including type 2 diabetes (T2D). In T2D, microbiome alterations are linked to metabolic dysregulation, insulin resistance, and chronic inflammation. However, many affected individuals also present with comorbid conditions or medication-related sensitivities and exposures, which may influence microbiome composition and complicate the interpretation of disease-associated microbial signatures. Despite the importance of these associations, microbiome sequencing data are challenging to analyze and interpret because of their large scale, heterogeneity, high dimensionality, and the complex relationships among microbial features. Conventional statistical approaches are limited in their ability to identify nonlinear patterns, multifactorial interactions, and disease-specific microbiome signatures. Moreover, many of these methods examine the role of potentially pathogenic factors primarily through linear associations, leaving substantial analytical gaps in the characterization of complex microbial contributions to disease. In this study, we use artificial intelligence–based modeling and available microbiome data from studies of T2D to provide a comparative interpretation of the disease-specific microbiome signature. The aim is to clarify the value and methodological role of SHAP in reanalyzing and interpreting microbiome data that have traditionally been evaluated using classical statistical approaches.

Keywords : human microbiome artificial intelligence machine learning shap microbial signatures chronic disease