A minimal 13-Transcript signature for HIV-TB coinfection in African children

Diagnosing tuberculosis (TB) in children living with HIV remains challenging due to difficulty in obtaining adequate samples, the paucibacillary nature of the disease, and symptom overlap with other infections. We previously identified hundreds of differentially expressed transcripts (DETs) in Ugandan and Southern African children living with HIV, yet their number limits clinical utility. Machine learning approaches can derive a compact, translatable diagnostic signature. We analyzed whole blood RNA-seq data from children living with HIV in Uganda (discovery, n = 59) and Southern Africa (Botswana, Eswatini; cross-regional test cohort, n = 38). An independent adult HIV-TB cohort from India (n = 25) provided an external test cohort. DETs were identified using DESeq2 and recursive feature elimination (RFE) with 10-fold cross-validation, and Random Forest was employed to derive a minimal signature. Performance was evaluated by AUC, sensitivity, and specificity. We derived a 13-transcript minimal signature from 566 DETs with a cross-validated AUC of 0.865 in the Ugandan cohort. In the Southern African cross-regional pediatric test cohort, AUC is 0.943 (95% CI 0.858–0.992), with all 16 NoTB cases correctly classified (specificity 1.000, Wilson 95% CI: 0.806–1.000). Performance in Indian adults is modest, with AUC of 0.773 (95% CI 0.564–0.929). Across cohorts, the signature outperforms neutrophil proportions alone. Nine transcripts converge on neutrophil degranulation. This 13-transcript signature, identified via machine learning, discriminates HIV-TB coinfection from HIV mono-infection in African children, with promising performance in a small Southern African cross-regional test cohort. The signature showed high specificity in African cohorts, though larger sample sizes are needed to establish its clinical utility as a rule-in test. Diagnosing tuberculosis (TB) in children with HIV is difficult because they often cannot provide the samples needed for standard tests, and their symptoms can look like other infections. In this study, we looked for an easier way to detect TB using a blood test. We had previously described how molecules called RNA transcripts differ between children with HIV and TB and those with HIV only. In this current study, we used advanced computer analysis to find the smallest number of RNA transcripts that can be used to distinguish children with HIV who have TB from those who do not. The signature correctly identified all children without TB in the Southern African test group, although this group was small. Nine of the 13 transcripts were linked to immune cells called neutrophils. These results suggest the signature may eventually help confirm TB in children with HIV, but validation in larger, diverse cohorts is essential before clinical use can be considered.