Predictive biomarkers for latent Mycobacterium tuberculosis infection

Abstract

Tuberculosis is a leading cause of infectious death worldwide, with almost a fourth of the world's population latently infected with its causative agent, Mycobacterium tuberculosis. Current diagnostic methods are insufficient to differentiate between healthy and latently infected populations. Here, we used a machine learning approach to analyze publicly available proteomic data from saliva and serum in Ethiopia's healthy, latent TB (LTBI) and active TB (ATBI) people. Our analysis discovered a profile of six proteins, Mast Cell Expressed Membrane Protein-1, Hemopexin, Lamin A/C, Small Proline Rich Protein 2F, Immunoglobulin Kappa Variable 4-1, and Voltage Dependent Anion Channel 2 that can precisely differentiate between the healthy and latently infected populations. This data suggests that a combination of six host proteins can serve as accurate biomarkers to diagnose latent infection. This is important for populations living in high-risk areas as it may help in the surveillance and prevention of severe disease.

Figure 1.. Protein biomarkers that predict Mtb latency with high accuracy against active TB and healthy controls.

A) Top 10 MCC values using single protein, two, three, and up to seven proteins via Random Forest prediction models. The Y-axis represents the MCC score, and X-axis represents the number of protein sets used to develop the model; B) Box plot showing the abundance values of five proteins with the highest MCC value. ANXA5, KRT6B, LCN2, ORM1, and MMP8. The Y-axis represents the relative abundance value of respective proteins in different cohorts.

Figure 2.. Classification scores of LTBI, PTB and control group in terms of percentage for each sample.

The default cutoff score of 0.5 was used to classify each sample. The cross sign represents a wrong prediction. The Y-axis represents the prediction score from each cohort.

Figure 3.. Protein biomarkers that predict Mtb latency with high accuracy against healthy controls.

A) Top 10 MCC values using single protein, pair of two, three, and up to seven proteins via Random Forest prediction models. The Y-axis represents the MCC score, and X-axis represents the number of protein sets used to develop the model. B) Box plot showing the abundance values of six proteins which has the highest MCC value. The Y-axis represents the relative abundance value of respective proteins in different cohorts.

Figure 4.. Classification scores of LTBI and control group in terms of percentage for each sample.

The default cutoff score of 0.5 was used to classify each sample. The cross sign represents the wrong prediction. The Y-axis represents the prediction score from each cohort.