Machine Learning-Driven Omics Approaches for Diagnostic Biomarker in Tuberculosis based on Differential Gene Expression data

Abstract

Background: Tuberculosis (TB) continues to remain a global health challenge, with 10.6 million cases and 1.3 million fatalities reported in 2022. Mycobacterium tuberculosis presents diagnostic difficulties because traditional procedures such as the Tuberculin Skin Test and IGRA are limited. MicroRNAs, including miR-18b and miR-378d, have been identified as prospective biomarkers for tuberculosis.

Method: This extensive review examines the use of OMICS technologies (genomics, transcriptomics, proteomics) with machine learning algorithms to identify and assess tuberculosis biomarkers. The recent literature from PubMed, Scopus, and IEEE Xplore was studied, concentrating on standard techniques such as KNN, SVM, and deep learning models.

Results: Machine learning techniques, especially deep learning, routinely attained elevated accuracy rates (often above 95%) in the classification of tuberculosis infections utilizing OMICS data. Biomarkers like miR-29a, miR-21, and 2-hydroxyglutarate (2-HG) have been recognized as promising diagnostic tools for TB. Additional biomarkers, including IL-8, IL-6, IFN-γ, and FCGR1A from transcriptomics, serum amyloid A (SAA), and lipoarabinomannan (LAM) from proteomics, provide more accurate insights into TB infection and progression.

Conclusion: The integration of machine learning with OMICS data offers a groundbreaking approach for tuberculosis diagnosis and biomarker discovery. Additional research is required to improve feature selection and refine machine learning models for clinical applications, potentially revolutionizing tuberculosis detection and treatment methods.