Maha Reda Farhat

Despite the long-held view of Mycobacterium tuberculosis ( Mtb ) as a genetically conserved pathogen, many genomic regions remain poorly resolved due to high sequence homology and repetitive content. Using complete genome assemblies generated from long-read sequencing of 151 globally representative clinical isolates, we comprehensively analyzed genome-wide patterns of genetic diversity and evolution across the Mtb genome. Our analysis uncovers pronounced diversity hotspots within paralogous regions (PRs) generated by recurrent gene conversion between homologous genes. In many cases, these hotspots exhibit more than an order of magnitude greater genetic diversity than the rest of the Mtb genome, which is otherwise characterized by remarkably low variation. Mutations within these regions display clustered substitution patterns, excess paralog-matching variants, and distinct mutational spectra consistent with ongoing gene conversion. Our analysis identifies over 300 individual gene conversion events (GCEs) distributed throughout the Mtb phylogeny. These GCEs occur predominantly within gene families associated with virulence and host–pathogen interactions, including the PE, PPE, and ESX families. Several of the most pronounced diversity hotspots occur in antigens encoded within PRs. Among these, the vaccine candidate PPE18 harbors mutations in validated epitope sequences and predicted alterations in HLA-II binding. Together, these findings demonstrate that gene conversion actively shapes antigenic and virulence-associated diversity in Mtb .

In Mycobacterium tuberculosis , a prevalent and deadly pathogen, resistance to antibiotics evolves primarily through non-synonymous mutations in proteins. Sequence-based analyses can uncover the genetic basis of antibiotic resistance, but these methods focus on primary sequence and often neglect other biological signals, such as protein structural information. We hypothesize that integrating the structural context of mutations improves the prediction of effects on function and phenotype. We curate high-confidence structural annotations for the M. tuberculosis proteome from 1350 crystallography and 2337 AlphaFold predictions, and mutations from over 31,000 M. tuberculosis isolates. We demonstrate that mutations in proteins known to cause resistance are clustered in 3D space, even in proteins where inactivating mutations at any position are thought to cause resistance. We find over 450 proteins in the M. tuberculosis proteome that display signals of clustered mutations, many of which have a known relationship with antibiotic resistance. We show that a supervised classifier trained on 3D distance to known resistance sites alone has an F 1 score of 96.5% at classifying mutations as resistance-conferring on a held-out test set. This work demonstrates that protein structure provides useful information for categorizing which variants may cause antibiotic resistance, even when the majority of structures are AI-predicted.

Insertions and deletions (INDELs) in simple sequence repeats (SSRs) generate relatively high-frequency reversible genetic changes that facilitate bacterial adaptation to changing environments. Analyses of global Mycobacterium tuberculosis ( Mtb ) isolates indicate that many SSRs are under diversifying selection, and several of the resulting INDELs in homopolymer tracts (HTs) can increase the pathogen’s fitness during exposure to host and antibiotic stresses. However, the functional impact of most variable SSRs, particularly those within more complex repeat sequences than HT, remains unclear. Here, we combine phylogenomic analysis of clinical Mtb strains from Vietnam and Peru with in vitro experimental validation of engineered strains to identify SSR INDELs that alter antibiotic susceptibility. Our findings demonstrate that INDELs across multiple SSRs of differing repeat composition are highly variable and correlate with clinical antibiotic resistance. These variants included frameshifting HT INDELs in ppe13 , glpK , Rv2081c, and ppsA , and in-frame trinucleotide (triplet) SSR INDELS in ponA1, ppe53, and ppe59 that produce much more subtle changes in protein structure. Reconstruction of these INDELs in an isogenic background identified four variants that directly reduce drug potency, including a triplet SSR deletion in ppe53 that conferred intermediate resistance to isoniazid, rifampicin, and streptomycin. The clinically prevalent ppe53 CGC _del mutation shortens a polyalanine stretch adjacent to the conserved WxG domain and impairs the processing and secretion of the full-length protein. Overall, our work provides additional evidence of selective pressure across Mtb SSRs and demonstrates the significance of in-frame INDELs within triplet SSRs, highlighting their contribution to the evolution of antibiotic resistance. Author Summary: Large scale analyses of genome sequences from clinical isolates of Mycobacterium tuberculosis have associated many mutations with altered antibiotic susceptibility, informing sequence-based prediction of treatment efficacy. However, previous studies often focus on single nucleotide polymorphisms, limiting our understanding of the roles of more complex mutagenic processes in the adaptation to drug pressure. In particular, insertions and deletions (INDELs) in repetitive regions are quite common, but their impact is incompletely understood. We show that INDELs in simple sequence repeats (SSRs), particularly in-frame INDELs within triplet SSRs, are not only associated with clinical resistance but also confer intermediate antibiotic resistance. In particular, a common one-codon deletion in the ppe53 gene shortens a polyalanine stretch in the encoded protein and appears to disrupt critical interactions with the secretion machine, thereby impairing processing and localization of the full-length protein. Our findings reveal a previously unappreciated mechanism of Mtb adaptation to antibiotic stress driven by in-frame mutations in triplet SSRs.

In Mycobacterium tuberculosis, a prevalent and deadly pathogen, resistance to antibiotics evolves primarily through non-synonymous mutations in proteins. Sequence-based analyses are currently used to understand the genetic basis of antibiotic resistance, either via genotype-phenotype association, or via signals of convergent evolution. These methods focus on primary sequence and often neglect other biological signals such as protein structural information. We hypothesize that integrating the structural context of mutations improves the prediction of effects on function and phenotype. We curate high confidence structural annotations for the M. tuberculosis proteome from 1,371 crystallography and 2,316 AlphaFold predictions, and combine the structures with mutations from over 31,000 clinical M. tuberculosis isolates. We demonstrate that mutations in proteins known to cause resistance are clustered in 3D space, even in proteins where inactivating mutations at any position are thought to cause resistance. We develop a statistic to search the M. tuberculosis proteome for signal of clustered mutations, finding over 450 proteins that display this signal, many of which have a known relationship with antibiotic resistance. We show that a supervised classifier trained on 3D distance to known resistance sites alone has an F1 score of 94.6% at classifying mutations as resistance-conferring across proteins. This work demonstrates that protein structure provides useful information for categorizing which variants may cause antibiotic resistance, even when the majority of structures are AI-predicted.

Machine learning (ML) has accelerated molecular discovery, yet training models to generalize to out-of-distribution (OOD) chemical spaces remains fundamentally constrained by the high cost of experimental validation. In antibiotic discovery, where whole-cell phenotypic high throughput screening (HTS) is resource-intensive, iterative ML-guided compound selection, or Active Learning (AL), offers a pathway to efficiently navigate available chemical spaces. However, the algorithmic tradeoffs between prioritizing compound novelty (exploration), predicted bioactivity (exploitation), and their impact on OOD generalizability remain unresolved for noisy, whole-cell biological systems. In this work, we systematically evaluate three AL strategies for whole-cell bacterial bioactivity and benchmark their effects on model accuracy, hit rate, and OOD performance. Using retrospective simulations on Mycobacterium tuberculosis HTS data, we identify an optimal AL strategy that balances predicted hit/non-hit novelty with overall hit rate. We then integrate the strategy in a closed-loop Borrelia burgdorferi antibiotic discovery HTS campaign. The AL-guided approach successfully increased the experimental screening hit rate five-fold (from a 0.2% rate within investigator-selected plates to 1.0%). Further, when the trained model was applied in prospective in silico selection of highly diverse compounds across multiple bacterial species, the AL-trained whole-cell inhibition predictor demonstrates 53-fold enrichment over investigator-directed screening (11.0% experimental validation of predicted hits). Of these, 100% demonstrated the intended narrow spectrum activity for Borrelia burgdorferi. These results demonstrate that calibrated AL strategies can overcome data acquisition bottlenecks and train generalizable property predictors able to extrapolate to OOD molecules.

Background: Tuberculosis (TB) treatment outcomes remain suboptimal, and standard clinical diagnostics cannot reliably identify patients at high risk of treatment failure or relapse at the time of diagnosis. While within-host Mycobacterium tuberculosis genetic diversity is hypothesized to reflect the viable bacterial burden and adaptive capacity of the infection, its clinical prognostic value remains unknown. Methods: We conducted a prospective cohort study of 364 patients with newly diagnosed, rifampicin-susceptible pulmonary TB in South Africa. Patients received standard 6-month therapy and were monitored for up to two years to ascertain composite unfavorable outcomes (treatment failure, death, or relapse). To accurately detect low-frequency (unfixed) genetic variants and eliminate reference bias artifacts, we mapped medium to high depth short-read sequences against matched, patient-specific long-read assemblies. The association between baseline pathogen genetic diversity and clinical outcomes was evaluated using multivariable Cox proportional-hazards models. Results: After bioinformatic filtering, true unfixed variants were relatively rare but significantly enriched in genes mediating pathogen adaptation and drug tolerance, including transporter proteins and two-component regulatory systems. Within-host bacterial genetic diversity ( i.e., the total number of unfixed variants) ranged from 0–20, with a median of 1 per patient. In survival analysis adjusting for known clinical risk factors—including HIV status, prior TB, baseline smear positivity, and radiographic lung involvement—baseline within-host genetic diversity emerged as a strong, independent predictor of unfavorable treatment outcomes. For patients with greater than 3 unfixed variants at diagnosis, each increase of 5 unfixed variants was associated with more than double the risk of a composite unfavorable outcome (adjusted Hazard Ratio, 2.36; 95% CI, 1.27 to 4.39; p=0.007). Conclusions: Baseline within-host pathogen genetic diversity is an independent predictor of unfavorable TB treatment outcomes. As sequencing becomes increasingly integrated into routine diagnostics, quantifying unfixed variants is an accessible approach that promises to risk-stratify patients and guide the duration of individualized regimens.

There is considerable interest in training machine learning (ML) models on genomic data that achieve clinical grade diagnostic accuracy. Many successful ML models have been trained and validated on binary tasks because predicting biomedically relevant continuous variables is difficult to optimize. In this work, we present convolutional neural networks (CNNs) that predict minimum inhibitory concentrations (MICs) for eight antibiotics from Mycobacterium tuberculosis complex (MTBC) gene sequences. By including evolutionary information, protein biochemical properties, and data augmentation for rare variants, we build models that predict 89% of MICs within one drug concentration doubling. Although trained on ≤ 52% of the World Health Organization’s (WHO) MTBC drug resistance mutation catalog data, the CNNs accurately predict the effects of 97% of the catalog’s graded mutations. In a cohort of 373 patients with rifampicin-susceptible M. tuberculosis infections, higher CNN-predicted rifampicin MICs are associated with unfavorable treatment outcomes, providing additional evidence that subtle differences in MIC below the resistance threshold are clinically relevant. These results demonstrate the value of encoding multiple dimensions of biological data in machine learning of M. tuberculosis drug resistance phenotypes and that domain knowledge-inspired machine learning models can be both interpretable and reach clinical grade accuracy.

Background: Chest X-ray (CXR) can assess pulmonary tuberculosis (TB) severity and may guide duration of treatment. However, the optimal radiological metric and its integration with clinical variables for predicting treatment outcomes remains unclear. Methods: We used logistic regression to associate human-read and commercial artificial intelligence–generated CXR metrics with unfavorable outcome in the TB Portals real-world dataset (n = 2809). We assessed the standalone predictive accuracy for each of 10 radiological features for unfavorable outcomes, and combined the best-performing features with other clinical data. We fine-tuned ensembles of convolutional neural nets (CNNs) to automate human-read percent of lung involved (PLI) measurement directly from CXR images (n = 5261). Results: Human-read PLI is the only CXR finding associated with outcome across drug resistance and human immunodeficiency virus (HIV) subgroups and is optimally combined with age, sex, and smear grade for predicting treatment outcome. PLI predicts outcomes better than cavitation (area under the curve [AUC], 0.654 vs 0.581, respectively), performs better than all tested Qure.ai automated features (qXR v2), and improves outcome prediction when added to sex + age + smear grade (ΔAUC, 0.028 [95% confidence interval, .007–.042]). The CNN ensemble for predicting PLI >25% achieves an AUC of 0.850. Conclusions: PLI performs better than cavitation as a radiological marker for predicting TB treatment outcome, and improves risk stratification when combined with key clinical variables. Automation of PLI can be predicted using CNNs to enable scalability and accurate assessment.

Tuberculosis (TB) is a subacute to chronic respiratory infection with insidious onset and protean symptoms. Treatment is complex and requires a multi-drug, multi-month regimen in which adherence is critical. Artificial intelligence (AI) offers promising solutions to challenges across the TB care cascade including screening, diagnosis and treatment. We conduct a scoping review of the literature published from 2017 to 2025 on the use of AI in TB screening, diagnosis, drug resistance diagnosis, treatment monitoring, and regimen design. We then extract data on study characteristics, AI methodology, input data, and sample size and describe AI tool performance and technology readiness level (TRL). Ninety studies are included, representing 803,383 study participants across 24 countries. Most studies (n=46) focus on radiological imaging for TB screening or diagnosis, but a burgeoning number of studies address drug resistance diagnosis (n=11), regimen design (n=4), treatment monitoring (n=12), and treatment adherence (n=8). Reported accuracy of AI interpretation of chest imaging for TB diagnosis was high at a median Area Under the receiver operating Curve (AUC) of 0.94 [IQR 0.12, range 0.81-0.99] for internal validation and 0.89 [IQR 0.14, range 0.66-0.98] for external validation, and a median TRL of 5 (IQR 1, range 4-7). AI demonstrates promise for advancing TB care towards World Health Organization (WHO) End TB targets, but several gaps remain, including AI-ready fit-for-task data availability, limited external validation and challenges in clinical integration. Closing these gaps will be critical for realizing the full potential of AI in TB care towards WHO End TB targets.

Despite the long-held view of Mycobacterium tuberculosis ( Mtb ) as a genetically conserved pathogen, many genomic regions remain poorly resolved due to high sequence homology and repetitive content. Using complete genome assemblies generated from long-read sequencing of 151 globally representative clinical isolates, we comprehensively analyzed genome-wide patterns of genetic diversity and evolution across the Mtb genome. Our analysis uncovers pronounced diversity hotspots within paralogous regions generated by recurrent gene conversion between homologous genes. In many cases, these hotspots exhibit more than an order of magnitude greater genetic diversity than the rest of the Mtb genome, which is otherwise characterized by remarkably low variation. Mutations within these regions display clustered substitution patterns, excess paralog-matching variants, and distinct mutational spectra consistent with ongoing gene conversion. Our analysis identifies over 300 individual gene conversion events distributed throughout the Mtb phylogeny. These gene conversion events occur predominantly within gene families associated with virulence and host-pathogen interactions, including the PE, PPE, and ESX families. Several of the most pronounced diversity hotspots occur in antigens encoded within paralogous regions. Among these, the vaccine candidate PPE18 harbors mutations in validated epitope sequences and predicted alterations in HLA-II binding. Together, these findings demonstrate that gene conversion actively shapes antigenic and virulence-associated diversity in Mtb .