Construction of a diagnostic model for tuberculosis based on long non-coding RNA

Area of Science:

• Biomarkers and Diagnostics
• Molecular Biology
• Computational Biology

Background:

• The World Health Organization advocates for new diagnostic tools for tuberculosis (TB) control, emphasizing non-sputum based methods.
• Current diagnostic approaches for TB often rely on sputum, which presents challenges in sample collection and processing.

Purpose of the Study:

• To develop and validate a machine learning-driven diagnostic model for TB using long non-coding RNAs (lncRNAs) as biomarkers.
• To establish a non-sputum based diagnostic approach for improved TB detection.

Main Methods:

• Microarray analysis of peripheral blood mononuclear cells (PBMCs) from TB patients, latent TB infection (LTBI) individuals, and healthy controls (HCs).
• Weighted gene co-expression network analysis (WGCNA) to identify TB-related lncRNA modules.
• Quantitative PCR (qPCR) for lncRNA validation, LASSO regression for feature selection, and AdaBoost algorithm for diagnostic model construction.

Main Results:

• WGCNA identified 12 key modules associated with TB; 14 differentially expressed lncRNAs were validated by qPCR.
• Five key lncRNAs were selected using LASSO regression for the diagnostic model.
• The AdaBoost model achieved an AUC of 0.97 in the training cohort and 0.91 in the validation cohort, with independent validation showing an AUC of 0.92.

Conclusions:

• A novel, blood-based diagnostic model incorporating five host-derived lncRNAs and an AdaBoost algorithm was established.
• This non-sputum diagnostic model demonstrates high accuracy and potential for enhancing TB diagnosis globally.