A three-methylation-driven gene-based deep learning model for tuberculosis diagnosis in patients with and without human immunodeficiency virus co-infection.

Abstract

Improved diagnostic tests for tuberculosis (TB) among people with human immunodeficiency virus (HIV) are urgently required. We hypothesized that methylation-driven genes (MDGs) of host blood could be used to diagnose patients co-infected with HIV/TB. In this study, we identified three MDGs among patients with HIV monoinfection and those with HIV/TB co-infection using the R package MethylMix. We then developed a deep learning model by screening these three MDGs, which distinguished HIV/TB co-infection from HIV monoinfection with a sensitivity of 95.2% and a specificity of 88.3%. On the two independent data sets, the sensitivity and specificity were 80%-92.8% and 72.7%-87.5%, respectively. Besides, our deep learning model accurately classified TB (sensitivity, 75.0%-100%; specificity, 91.3%-98.1%) and other respiratory disorders (sensitivity, 72.7%-75.0%; specificity, 70.9%-72.7%). This study will contribute to improve molecular diagnosis for HIV/TB co-infection.