Percent of lung involved in disease on chest X-ray predicts unfavorable treatment outcome in pulmonary tuberculosis.

Radiology may better define tuberculosis (TB) severity and guide duration of treatment. We aimed to systematically study baseline chest X-rays (CXR) and their association with TB treatment outcome using real-world data. We used logistic regression to associate TB treatment outcomes with CXR findings, including percent of lung involved in disease (PLI), cavitation, and Timika score, alone or in combination with other clinical characteristics, stratifying by drug resistance status and HIV (n = 2,809). We fine-tuned convolutional neural nets (CNN) to automate PLI measurement from the CXR DICOM images (n = 5,261). PLI is the only CXR finding associated with unfavorable outcome across drug resistance and HIV subgroups [Rifampicin-susceptible disease without HIV, adjusted odds ratio (aOR) 1·11 (1·01, 1·22), P-value 0·025]. The most informed model of baseline characteristics tested predicts outcome with a validation mean area under the curve (AUC) of 0·769. PLI and Timika (AUC 0·656 and 0·655 respectively) predict unfavorable outcomes better than cavitary information (best AUC 0·591). The addition of PLI improves prediction compared to sex and age alone (AUC 0·680 and 0·627, respectively).PLI>25% provides a better separation of favorable and unfavorable outcomes compared to PLI>50%. The best performing ensemble of CNNs has an AUC 0·850 for PLI>25% and mean absolute error of 11·7% for the PLI value. PLI is better than cavitation for predicting unfavorable treatment outcome in pulmonary TB in non-clinical trial settings and it can be accurately and automatically predicted with CNNs. One Sentence Summary: The percent of lung involved in disease improves prediction of unfavorable outcomes in pulmonary tuberculosis when added to clinical characteristics.

Differential rates of Mycobacterium tuberculosis transmission associate with host-pathogen sympatry.

Several human-adapted Mycobacterium tuberculosis complex (Mtbc) lineages exhibit a restricted geographical distribution globally. These lineages are hypothesized to transmit more effectively among sympatric hosts, that is, those that share the same geographical area, though this is yet to be confirmed while controlling for exposure, social networks and disease risk after exposure. Using pathogen genomic and contact tracing data from 2,279 tuberculosis cases linked to 12,749 contacts from three low-incidence cities, we show that geographically restricted Mtbc lineages were less transmissible than lineages that have a widespread global distribution. Allopatric host-pathogen exposure, in which the restricted pathogen and host are from non-overlapping areas, had a 38% decrease in the odds of infection among contacts compared with sympatric exposures. We measure tenfold lower uptake of geographically restricted lineage 6 strains compared with widespread lineage 4 strains in allopatric macrophage infections. We conclude that Mtbc strain-human long-term coexistence has resulted in differential transmissibility of Mtbc lineages and that this differs by human population.

Beyond diagnostic connectivity: Leveraging digital health technology for the real-time collection and provision of high-quality actionable data on infectious diseases in Uganda.

Automated data transmission from diagnostic instrument networks to a central database at the Ministries of Health has the potential of providing real-time quality data not only on diagnostic instrument performance, but also continuous disease surveillance and patient care. We aimed at sharing how a locally developed novel diagnostic connectivity solution channels actionable data from diagnostic instruments to the national dashboards for disease control in Uganda between May 2022 and May 2023. The diagnostic connectivity solution was successfully configured on a selected network of multiplexing diagnostic instruments at 260 sites in Uganda, providing a layered access of data. Of these, 909,674 test results were automatically collected from 269 "GeneXpert" machines, 5597 test results from 28 "Truenat" and >12,000 were from 3 digital x-ray devices to different stakeholder levels to ensure optimal use of data for their intended purpose. The government and relevant stakeholders are empowered with usable and actionable data from the diagnostic instruments. The successful implementation of the diagnostic connectivity solution depended on some key operational strategies namely; sustained internet connectivity and short message services, stakeholder engagement, a strong in-country laboratory coordination network, human resource capacity building, establishing a network for the diagnostic instruments, and integration with existing health data collection tools. Poor bandwidth at some locations was a major hindrance for the successful implementation of the connectivity solution. Maintaining stakeholder engagement at the clinical level is key for sustaining diagnostic data connectivity. The locally developed diagnostic connectivity solution as a digital health technology offers the chance to collect high-quality data on a number of parameters for disease control, including error analysis, thereby strengthening the quality of data from the networked diagnostic sites to relevant stakeholders.

Exploring gene content with pangene graphs.

MOTIVATION: The gene content regulates the biology of an organism. It varies between species and between individuals of the same species. Although tools have been developed to identify gene content changes in bacterial genomes, none is applicable to collections of large eukaryotic genomes such as the human pangenome. RESULTS: We developed pangene, a computational tool to identify gene orientation, gene order and gene copy-number changes in a collection of genomes. Pangene aligns a set of input protein sequences to the genomes, resolves redundancies between protein sequences and constructs a gene graph with each genome represented as a walk in the graph. It additionally finds subgraphs, which we call bibubbles, that capture gene content changes. Applied to the human pangenome, pangene identifies known gene-level variations and reveals complex haplotypes that are not well studied before. Pangene also works with high-quality bacterial pangenome and reports similar numbers of core and accessory genes in comparison to existing tools. AVAILABILITY AND IMPLEMENTATION: Source code at https://github.com/lh3/pangene; pre-built pangene graphs can be downloaded from https://zenodo.org/records/8118576 and visualized at https://pangene.bioinweb.org.

MmpR5 protein truncation and bedaquiline resistance in Mycobacterium tuberculosis isolates from South Africa: a genomic analysis.

BACKGROUND: The antibiotic bedaquiline is a key component of new WHO regimens for drug-resistant tuberculosis; however, predicting bedaquiline resistance from bacterial genotypes remains challenging. We aimed to understand the genetic mechanisms of bedaquiline resistance by analysing Mycobacterium tuberculosis isolates from South Africa. METHODS: For this genomic analysis, we conducted whole-genome sequencing of Mycobacterium tuberculosis samples collected at two referral laboratories in Cape Town and Johannesburg, covering regions of South Africa with a high prevalence of tuberculosis. We used the tool ARIBA to measure the status of predefined genes that are associated with bedaquiline resistance. To produce a broad genetic landscape of M tuberculosis in South Africa, we extended our analysis to include all publicly available isolates from the European Nucleotide Archive, including isolates obtained by the CRyPTIC consortium, for which minimum inhibitory concentrations of bedaquiline were available. FINDINGS: Between Jan 10, 2019, and July, 22, 2020, we sequenced 505 M tuberculosis isolates from 461 patients. Of the 64 isolates with mutations within the mmpR5 regulatory gene, we found 53 (83%) had independent acquisition of 31 different mutations, with a particular enrichment of truncated MmpR5 in bedaquiline-resistant isolates resulting from either frameshift mutations or the introduction of an insertion element. Truncation occurred across three M tuberculosis lineages, and were present in 66% of bedaquiline-resistant isolates. Although the distributions overlapped, the median minimum inhibitory concentration of bedaquiline was 0·25 mg/L (IQR 0·12-0·25) in mmpR5-disrupted isolates, compared with 0·06 mg/L (0·03-0·06) in wild-type M tuberculosis. INTERPRETATION: Reduction in the susceptibility of M tuberculosis to bedaquiline has evolved repeatedly across the phylogeny. In our data, we see no evidence that this reduction has led to the spread of a successful strain in South Africa. Binary phenotyping based on the bedaquiline breakpoint might be inappropriate to monitor resistance to this drug. We recommend the use of minimum inhibitory concentrations in addition to MmpR5 truncation screening to identify moderate increases in resistance to bedaquiline. FUNDING: US Centers for Disease Control and Prevention.

Single nucleotide variation catalog from clinical isolates mapped on tertiary and quaternary structures of ESX-1-related proteins reveals critical regions as putative Mtb therapeutic targets.

Proteins encoded by the ESX-1 genes of interest are essential for full virulence in all Mycobacterium tuberculosis complex (Mtbc) lineages, the pathogens causing the highest mortality worldwide. Identifying critical regions in these ESX-1-related proteins could provide preventive or therapeutic targets for Mtb infection, the game changer needed for tuberculosis control. We analyzed a compendium of whole genome sequences of clinical Mtb isolates from all lineages from >32,000 patients and identified single nucleotide polymorphisms. When mutations corresponding to all non-synonymous single nucleotide polymorphisms were mapped on structural models of the ESX-1 proteins, fully conserved regions emerged. Some could be assigned to known quaternary structures, whereas others could be predicted to be involved in yet-to-be-discovered interactions. Some mutants had clonally expanded (found in >1% of the isolates); these mutants were mostly located at the surface of globular domains, remote from known intra- and inter-molecular protein-protein interactions. Fully conserved intrinsically disordered regions of proteins were found, suggesting that these regions are crucial for the pathogenicity of the Mtbc. Altogether, our findings highlight fully conserved regions of proteins as attractive vaccine antigens and drug targets to control Mtb virulence. Extending this approach to the whole Mtb genome as well as other microorganisms will enhance vaccine development for various pathogens. IMPORTANCE: We mapped all non-synonymous single nucleotide polymorphisms onto each of the experimental and predicted ESX-1 proteins' structural models and inspected their placement. Varying sizes of conserved regions were found. Next, we analyzed predicted intrinsically disordered regions within our set of proteins, finding two putative long stretches that are fully conserved, and discussed their potential essential role in immunological recognition. Combined, our findings highlight new targets for interfering with Mycobacterium tuberculosis complex virulence.

Drug resistance and epidemiological success of modern Mycobacterium tuberculosis lineages in western India.

BACKGROUND: Transmission is contributing to the slow decline of tuberculosis (TB) incidence globally. Drivers of TB transmission in India, the country estimated to carry a quarter of the World's burden, are not well studied. We conducted a genomic epidemiology study to compare epidemiological success, host factors and drug resistance (DR) among the four major Mycobacterium tuberculosis (Mtb) lineages (L1-4) circulating in Pune, India. METHODS: We performed whole-genome sequencing (WGS) of Mtb sputum culture-positive isolates from participants in two prospective cohort studies and predicted genotypic susceptibility using a validated random forest model. We used maximum likelihood estimation to build phylogenies. We compared lineage specific phylogenetic and time-scaled metrics to assess epidemiological success. RESULTS: Of the 642 isolates that underwent WGS, 612 met sequence quality criteria. Most isolates belonged to L3 (44.6%). The majority (61.1%) of multidrug-resistant isolates belonged to L2 (P < 0.001). In molecular dating, L2 demonstrated a higher rate and more recent resistance acquisition. We measured higher clustering, and time-scaled haplotypic density (THD) for L4 and L2 compared to L3 and/or L1 suggesting higher epidemiological success. L4 demonstrated higher THD and clustering (OR 5.1 (95% CI 2.3-12.3) in multivariate models controlling for host factors and DR. CONCLUSION: L2 shows a higher frequency of DR and both L2 and L4 demonstrate evidence of higher epidemiological success than L3 or L1 in the study setting. Our findings highlight the need for contact tracing around TB cases, and heightened surveillance of TB DR in India.

Identification of bacterial determinants of tuberculosis infection and treatment outcomes: a phenogenomic analysis of clinical strains.

BACKGROUND: Bacterial diversity could contribute to the diversity of tuberculosis infection and treatment outcomes observed clinically, but the biological basis of this association is poorly understood. The aim of this study was to identify associations between phenogenomic variation in Mycobacterium tuberculosis and tuberculosis clinical features. METHODS: We developed a high-throughput platform to define phenotype-genotype relationships in M tuberculosis clinical isolates, which we tested on a set of 158 drug-sensitive M tuberculosis strains sampled from a large tuberculosis clinical study in Ho Chi Minh City, Viet Nam. We tagged the strains with unique genetic barcodes in multiplicate, allowing us to pool the strains for in-vitro competitive fitness assays across 16 host-relevant antibiotic and metabolic conditions. Relative fitness was quantified by deep sequencing, enumerating output barcode read counts relative to input normalised values. We performed a genome-wide association study to identify phylogenetically linked and monogenic mutations associated with the in-vitro fitness phenotypes. These genetic determinants were further associated with relevant clinical outcomes (cavitary disease and treatment failure) by calculating odds ratios (ORs) with binomial logistic regressions. We also assessed the population-level transmission of strains associated with cavitary disease and treatment failure using terminal branch length analysis of the phylogenetic data. FINDINGS: M tuberculosis clinical strains had diverse growth characteristics in host-like metabolic and drug conditions. These fitness phenotypes were highly heritable, and we identified monogenic and phylogenetically linked variants associated with the fitness phenotypes. These data enabled us to define two genetic features that were associated with clinical outcomes. First, mutations in Rv1339, a phosphodiesterase, which were associated with slow growth in glycerol, were further associated with treatment failure (OR 5·34, 95% CI 1·21-23·58, p=0·027). Second, we identified a phenotypically distinct slow-growing subclade of lineage 1 strains (L1.1.1.1) that was associated with cavitary disease (OR 2·49, 1·11-5·59, p=0·027) and treatment failure (OR 4·76, 1·53-14·78, p=0·0069), and which had shorter terminal branch lengths on the phylogenetic tree, suggesting increased transmission. INTERPRETATION: Slow growth under various antibiotic and metabolic conditions served as in-vitro intermediate phenotypes underlying the association between M tuberculosis monogenic and phylogenetically linked mutations and outcomes such as cavitary disease, treatment failure, and transmission potential. These data suggest that M tuberculosis growth regulation is an adaptive advantage for bacterial success in human populations, at least in some circumstances. These data further suggest markers for the underlying bacterial processes that contribute to these clinical outcomes. FUNDING: National Health and Medical Research Council/A∗STAR, National Institutes of Allergy and Infectious Diseases, National Institute of Child Health and Human Development, and the Wellcome Trust Fellowship in Public Health and Tropical Medicine.

Precision Phenotyping for Curating Research Cohorts of Patients with Post-Acute Sequelae of COVID-19 (PASC) as a Diagnosis of Exclusion.

Scalable identification of patients with the post-acute sequelae of COVID-19 (PASC) is challenging due to a lack of reproducible precision phenotyping algorithms and the suboptimal accuracy, demographic biases, and underestimation of the PASC diagnosis code (ICD-10 U09.9). In a retrospective case-control study, we developed a precision phenotyping algorithm for identifying research cohorts of PASC patients, defined as a diagnosis of exclusion. We used longitudinal electronic health records (EHR) data from over 295 thousand patients from 14 hospitals and 20 community health centers in Massachusetts. The algorithm employs an attention mechanism to exclude sequelae that prior conditions can explain. We performed independent chart reviews to tune and validate our precision phenotyping algorithm. Our PASC phenotyping algorithm improves precision and prevalence estimation and reduces bias in identifying Long COVID patients compared to the U09.9 diagnosis code. Our algorithm identified a PASC research cohort of over 24 thousand patients (compared to about 6 thousand when using the U09.9 diagnosis code), with a 79.9 percent precision (compared to 77.8 percent from the U09.9 diagnosis code). Our estimated prevalence of PASC was 22.8 percent, which is close to the national estimates for the region. We also provide an in-depth analysis outlining the clinical attributes, encompassing identified lingering effects by organ, comorbidity profiles, and temporal differences in the risk of PASC. The PASC phenotyping method presented in this study boasts superior precision, accurately gauges the prevalence of PASC without underestimating it, and exhibits less bias in pinpointing Long COVID patients. The PASC cohort derived from our algorithm will serve as a springboard for delving into Long COVID's genetic, metabolomic, and clinical intricacies, surmounting the constraints of recent PASC cohort studies, which were hampered by their limited size and available outcome data.

Implications of Race Adjustment in Lung-Function Equations.

BACKGROUND: Adjustment for race is discouraged in lung-function testing, but the implications of adopting race-neutral equations have not been comprehensively quantified. METHODS: We obtained longitudinal data from 369,077 participants in the National Health and Nutrition Examination Survey, U.K. Biobank, the Multi-Ethnic Study of Atherosclerosis, and the Organ Procurement and Transplantation Network. Using these data, we compared the race-based 2012 Global Lung Function Initiative (GLI-2012) equations with race-neutral equations introduced in 2022 (GLI-Global). Evaluated outcomes included national projections of clinical, occupational, and financial reclassifications; individual lung-allocation scores for transplantation priority; and concordance statistics (C statistics) for clinical prediction tasks. RESULTS: Among the 249 million persons in the United States between 6 and 79 years of age who are able to produce high-quality spirometric results, the use of GLI-Global equations may reclassify ventilatory impairment for 12.5 million persons, medical impairment ratings for 8.16 million, occupational eligibility for 2.28 million, grading of chronic obstructive pulmonary disease for 2.05 million, and military disability compensation for 413,000. These potential changes differed according to race; for example, classifications of nonobstructive ventilatory impairment may change dramatically, increasing 141% (95% confidence interval [CI], 113 to 169) among Black persons and decreasing 69% (95% CI, 63 to 74) among White persons. Annual disability payments may increase by more than $1 billion among Black veterans and decrease by $0.5 billion among White veterans. GLI-2012 and GLI-Global equations had similar discriminative accuracy with regard to respiratory symptoms, health care utilization, new-onset disease, death from any cause, death related to respiratory disease, and death among persons on a transplant waiting list, with differences in C statistics ranging from -0.008 to 0.011. CONCLUSIONS: The use of race-based and race-neutral equations generated similarly accurate predictions of respiratory outcomes but assigned different disease classifications, occupational eligibility, and disability compensation for millions of persons, with effects diverging according to race. (Funded by the National Heart Lung and Blood Institute and the National Institute of Environmental Health Sciences.).

Analysis of the limited M. tuberculosis accessory genome reveals potential pitfalls of pan-genome analysis approaches.

Pan-genome analysis is a fundamental tool for studying bacterial genome evolution; however, the variety of methods used to define and measure the pan-genome poses challenges to the interpretation and reliability of results. To quantify sources of bias and error related to common pan-genome analysis approaches, we evaluated different approaches applied to curated collection of 151 Mycobacterium tuberculosis ( Mtb ) isolates. Mtb is characterized by its clonal evolution, absence of horizontal gene transfer, and limited accessory genome, making it an ideal test case for this study. Using a state-of-the-art graph-genome approach, we found that a majority of the structural variation observed in Mtb originates from rearrangement, deletion, and duplication of redundant nucleotide sequences. In contrast, we found that pan-genome analyses that focus on comparison of coding sequences (at the amino acid level) can yield surprisingly variable results, driven by differences in assembly quality and the softwares used. Upon closer inspection, we found that coding sequence annotation discrepancies were a major contributor to inflated Mtb accessory genome estimates. To address this, we developed panqc, a software that detects annotation discrepancies and collapses nucleotide redundancy in pan-genome estimates. When applied to Mtb and E. coli pan-genomes, panqc exposed distinct biases influenced by the genomic diversity of the population studied. Our findings underscore the need for careful methodological selection and quality control to accurately map the evolutionary dynamics of a bacterial species.

For Tuberculosis, Not "To Screen or Not to Screen?" but "Who?" and "How?"

Active case finding leveraging new molecular diagnostics and chest X-rays with automated interpretation algorithms is increasingly being developed for high-risk populations to drive down tuberculosis incidence. We consider why such an approach did not deliver a decline in tuberculosis prevalence in Brazilian prison populations and what to consider next.

Transcriptomic responses to antibiotic exposure in Mycobacterium tuberculosis.

Transcriptional responses in bacteria following antibiotic exposure offer insights into antibiotic mechanism of action, bacterial responses, and characterization of antimicrobial resistance. We aimed to define the transcriptional antibiotic response (TAR) in Mycobacterium tuberculosis (Mtb) isolates for clinically relevant drugs by pooling and analyzing Mtb microarray and RNA-seq data sets. We generated 99 antibiotic transcription profiles across 17 antibiotics, with 76% of profiles generated using 3-24 hours of antibiotic exposure and 49% within one doubling of the WHO antibiotic critical concentration. TAR genes were time-dependent, and largely specific to the antibiotic mechanism of action. TAR signatures performed well at predicting antibiotic exposure, with the area under the receiver operating curve (AUC) ranging from 0.84-1.00 (TAR <6 hours of antibiotic exposure) and 0.76-1.00 (>6 hours of antibiotic exposure) for upregulated genes and 0.57-0.90 and 0.87-1.00, respectfully, for downregulated genes. This work desmonstrates that transcriptomics allows for the assessment of antibiotic activity in Mtb within 6 hours of exposure.

Drug-resistant tuberculosis: a persistent global health concern.

Drug-resistant tuberculosis (TB) is estimated to cause 13% of all antimicrobial resistance-attributable deaths worldwide and is driven by both ongoing resistance acquisition and person-to-person transmission. Poor outcomes are exacerbated by late diagnosis and inadequate access to effective treatment. Advances in rapid molecular testing have recently improved the diagnosis of TB and drug resistance. Next-generation sequencing of Mycobacterium tuberculosis has increased our understanding of genetic resistance mechanisms and can now detect mutations associated with resistance phenotypes. All-oral, shorter drug regimens that can achieve high cure rates of drug-resistant TB within 6-9 months are now available and recommended but have yet to be scaled to global clinical use. Promising regimens for the prevention of drug-resistant TB among high-risk contacts are supported by early clinical trial data but final results are pending. A person-centred approach is crucial in managing drug-resistant TB to reduce the risk of poor treatment outcomes, side effects, stigma and mental health burden associated with the diagnosis. In this Review, we describe current surveillance of drug-resistant TB and the causes, risk factors and determinants of drug resistance as well as the stigma and mental health considerations associated with it. We discuss recent advances in diagnostics and drug-susceptibility testing and outline the progress in developing better treatment and preventive therapies.

Exploring gene content with pangene graphs.

Motivation: The gene content regulates the biology of an organism. It varies between species and between individuals of the same species. Although tools have been developed to identify gene content changes in bacterial genomes, none is applicable to collections of large eukaryotic genomes such as the human pangenome. Results: We developed pangene, a computational tool to identify gene orientation, gene order and gene copy-number changes in a collection of genomes. Pangene aligns a set of input protein sequences to the genomes, resolves redundancies between protein sequences and constructs a gene graph with each genome represented as a walk in the graph. It additionally finds subgraphs, which we call bibubbles, that capture gene content changes. Applied to the human pangenome, pangene identifies known gene-level variations and reveals complex haplotypes that are not well studied before. Pangene also works with high-quality bacterial pangenome and reports similar numbers of core and accessory genes in comparison to existing tools. Availability and implementation: Source code at https://github.com/lh3/pangene; pre-built pangene graphs can be downloaded from https://zenodo.org/records/8118576 and visualized at https://pangene.bioinweb.org.

Estimation of country-specific tuberculosis resistance antibiograms using pathogen genomics and machine learning.

BACKGROUND: Global tuberculosis (TB) drug resistance (DR) surveillance focuses on rifampicin. We examined the potential of public and surveillance Mycobacterium tuberculosis (Mtb) whole-genome sequencing (WGS) data, to generate expanded country-level resistance prevalence estimates (antibiograms) using in silico resistance prediction. METHODS: We curated and quality-controlled Mtb WGS data. We used a validated random forest model to predict phenotypic resistance to 12 drugs and bias-corrected for model performance, outbreak sampling and rifampicin resistance oversampling. Validation leveraged a national DR survey conducted in South Africa. RESULTS: Mtb isolates from 29 countries (n=19 149) met sequence quality criteria. Global marginal genotypic resistance among mono-resistant TB estimates overlapped with the South African DR survey, except for isoniazid, ethionamide and second-line injectables, which were underestimated (n=3134). Among multidrug resistant (MDR) TB (n=268), estimates overlapped for the fluoroquinolones but overestimated other drugs. Globally pooled mono-resistance to isoniazid was 10.9% (95% CI: 10.2-11.7%, n=14 012). Mono-levofloxacin resistance rates were highest in South Asia (Pakistan 3.4% (0.1-11%), n=111 and India 2.8% (0.08-9.4%), n=114). Given the recent interest in drugs enhancing ethionamide activity and their expected activity against isolates with resistance discordance between isoniazid and ethionamide, we measured this rate and found it to be high at 74.4% (IQR: 64.5-79.7%) of isoniazid-resistant isolates predicted to be ethionamide susceptible. The global susceptibility rate to pyrazinamide and levofloxacin among MDR was 15.1% (95% CI: 10.2-19.9%, n=3964). CONCLUSIONS: This is the first attempt at global Mtb antibiogram estimation. DR prevalence in Mtb can be reliably estimated using public WGS and phenotypic resistance prediction for key antibiotics, but public WGS data demonstrates oversampling of isolates with higher resistance levels than MDR. Nevertheless, our results raise concerns about the empiric use of short-course fluoroquinolone regimens for drug-susceptible TB in South Asia and indicate underutilisation of ethionamide in MDR treatment.

Evaluating generalizability of artificial intelligence models for molecular datasets.

Deep learning has made rapid advances in modeling molecular sequencing data. Despite achieving high performance on benchmarks, it remains unclear to what extent deep learning models learn general principles and generalize to previously unseen sequences. Benchmarks traditionally interrogate model generalizability by generating metadata based (MB) or sequence-similarity based (SB) train and test splits of input data before assessing model performance. Here, we show that this approach mischaracterizes model generalizability by failing to consider the full spectrum of cross-split overlap, i.e., similarity between train and test splits. We introduce Spectra, a spectral framework for comprehensive model evaluation. For a given model and input data, Spectra plots model performance as a function of decreasing cross-split overlap and reports the area under this curve as a measure of generalizability. We apply Spectra to 18 sequencing datasets with associated phenotypes ranging from antibiotic resistance in tuberculosis to protein-ligand binding to evaluate the generalizability of 19 state-of-the-art deep learning models, including large language models, graph neural networks, diffusion models, and convolutional neural networks. We show that SB and MB splits provide an incomplete assessment of model generalizability. With Spectra, we find as cross-split overlap decreases, deep learning models consistently exhibit a reduction in performance in a task- and model-dependent manner. Although no model consistently achieved the highest performance across all tasks, we show that deep learning models can generalize to previously unseen sequences on specific tasks. Spectra paves the way toward a better understanding of how foundation models generalize in biology.

Genotype-Phenotype Characterization of Serial Mycobacterium tuberculosis Isolates in Bedaquiline-Resistant Tuberculosis.

BACKGROUND: Emerging resistance to bedaquiline (BDQ) threatens to undermine advances in the treatment of drug-resistant tuberculosis (DRTB). Characterizing serial Mycobacterium tuberculosis (Mtb) isolates collected during BDQ-based treatment can provide insights into the etiologies of BDQ resistance in this important group of DRTB patients. METHODS: We measured mycobacteria growth indicator tube (MGIT)-based BDQ minimum inhibitory concentrations (MICs) of Mtb isolates collected from 195 individuals with no prior BDQ exposure who were receiving BDQ-based treatment for DRTB. We conducted whole-genome sequencing on serial Mtb isolates from all participants who had any isolate with a BDQ MIC >1 collected before or after starting treatment (95 total Mtb isolates from 24 participants). RESULTS: Sixteen of 24 participants had BDQ-resistant TB (MGIT MIC ≥4 µg/mL) and 8 had BDQ-intermediate infections (MGIT MIC = 2 µg/mL). Participants with pre-existing resistance outnumbered those with resistance acquired during treatment, and 8 of 24 participants had polyclonal infections. BDQ resistance was observed across multiple Mtb strain types and involved a diverse catalog of mmpR5 (Rv0678) mutations, but no mutations in atpE or pepQ. Nine pairs of participants shared genetically similar isolates separated by <5 single nucleotide polymorphisms, concerning for potential transmitted BDQ resistance. CONCLUSIONS: BDQ-resistant TB can arise via multiple, overlapping processes, including transmission of strains with pre-existing resistance. Capturing the within-host diversity of these infections could potentially improve clinical diagnosis, population-level surveillance, and molecular diagnostic test development.

Multimodal learning with graphs.

Artificial intelligence for graphs has achieved remarkable success in modeling complex systems, ranging from dynamic networks in biology to interacting particle systems in physics. However, the increasingly heterogeneous graph datasets call for multimodal methods that can combine different inductive biases-the set of assumptions that algorithms use to make predictions for inputs they have not encountered during training. Learning on multimodal datasets presents fundamental challenges because the inductive biases can vary by data modality and graphs might not be explicitly given in the input. To address these challenges, multimodal graph AI methods combine different modalities while leveraging cross-modal dependencies using graphs. Diverse datasets are combined using graphs and fed into sophisticated multimodal architectures, specified as image-intensive, knowledge-grounded and language-intensive models. Using this categorization, we introduce a blueprint for multimodal graph learning, use it to study existing methods and provide guidelines to design new models.