Project description:Background: Tuberculosis (TB) among persons living with HIV (PLWH) poses diagnostic challenges. Although several transcriptional signatures have recently been identified as promising tool for TB diagnosis, data are limited in persons with advanced HIV. Methodology: Reads were aligned to the human transcriptome (GRCh38 version 100), comprising both mRNA and ncRNA, with Salmon v1.2.0. A Random Forest algorithm with “leave-one-out” cross-validation was applied in the variance stabilizing transformation gene expression. The identified biomarkers were compared to previous TB transcriptional signatures using the Area Under the Curve (AUC). Results: Functional analysis indicated that common upregulated pathways in TB/HIV patients were associated with Toll-like receptor cascades and neutrophil degranulation. A machine learning decision tree algorithm identified the expression values from RAB20 and INSL3 as most informative for classifying TB status in PLWH. Only these two genes were able to correctly classify all samples.