A predictive model for evaluating the risk of latent tuberculosis relapse via machine learning

Area of Science:

• Genomics
• Immunology
• Infectious Disease

Background:

• Reactivation of latent tuberculosis infection (LTBI) hinders tuberculosis eradication efforts.
• Predicting LTBI relapse is critical for effective disease management but is currently underexplored.

Purpose of the Study:

• To identify genes associated with LTBI relapse.
• To develop a predictive model for LTBI reactivation risk.

Main Methods:

• Differential gene expression analysis and WGCNA were used on GSE54992 data.
• A predictive model was built using SVM-RFE, LASSO, and random forest algorithms, selecting ALG2, FARS2, and PGP.
• Model performance was evaluated using AUC, and gene associations with immune pathways were explored via GSEA.

Main Results:

• 279 co-differentially expressed genes were identified between LTBI and active TB infection (ATBI) patients.
• The three-gene model (ALG2, FARS2, PGP) achieved AUC values of 0.781 and 0.725 in validation.
• ALG2 correlated with immune responses, FARS2 and PGP with metabolism; all three correlated with BCG load.

Conclusions:

• An integrative model using ALG2, FARS2, and PGP offers a robust tool for assessing LTBI reactivation risk.
• This model supports personalized strategies for tuberculosis management.