Rapid and accurate diagnostics of bacterial infections are necessary for efficient treatment of antibiotic-resistant pathogens. Cultivation-based methods, such as antibiotic susceptibility testing (AST), are limited by bacterial growth rates and seldom yield results before treatment needs to start, increasing patient risk and contributing to antibiotic overprescription. Here, we present a deep-learning method that leverages patient data and available AST results to predict antibiotic susceptibilities that have not yet been measured. After training on three million AST results from 30 European countries, the method achieved an average accuracy of 93% across bacterial species and antibiotics. It predicted susceptibility with an average major error rate below 5% for quinolones, cephalosporins, and carbapenems, and below 8% and 14% for aminoglycosides and penicillins, respectively. Furthermore, the model predicted resistance with an average very major error rate below 10% for cephalosporins, carbapenems, and aminoglycosides, but with higher very major error rates for penicillins and quinolones. We combined the method with conformal prediction and demonstrated accurate estimation of the predictive uncertainty at the patient level. Our results suggest that artificial intelligence-based decision support may offer new means to meet the growing burden of antibiotic resistance.IMPORTANCEImproved diagnostic tools are vital for maintaining efficient treatment of antibiotic-resistant bacteria and for reducing antibiotic overconsumption. In our research, we introduce a new deep learning-based method capable of predicting untested antibiotic resistance phenotypes. The method uses transformers, a powerful artificial intelligence (AI) technique that efficiently leverages both antibiotic susceptibility tests (AST) and patient data simultaneously. The model produces predictions that can be used as time- and cost-efficient alternatives to results from cultivation-based diagnostic assays. Significantly, our study highlights the potential of AI technologies to address the increasing prevalence of antibiotic-resistant bacterial infections.

Abstract Antimicrobial resistance is a public health challenge, driving the need for rapid, cost-effective diagnostic support tools. Artificial intelligence (AI) may enable prediction of susceptibility to untested antibiotics from known susceptibility results, but prospective clinical validation is required before routine use. We evaluated an AI-based decision support method, trained on invasive isolates from the European Surveillance System (TESSy), for prediction of antibiotic susceptibility in clinical Escherichia coli urine isolates. The evaluation included 99 E. coli isolates from urine samples with diversity in age, sex, and antibiotic susceptibility. Predictions were evaluated for 14 antibiotics using patient metadata and susceptibility results for 4–8 antibiotics as input. Prediction uncertainty was handled using conformal prediction, allowing abstention when confidence was insufficient. EUCAST disk diffusion test results were used as reference and genomic sequence data was used to explore mechanisms of the AI performance. Without conformal prediction, 84% of predictions were correct when susceptibility results of six antibiotics were used to predict susceptibility to eight additional antibiotics. Across all predictions generated using susceptibility results for six antibiotics as input, the major and very major error rates were 19% and 12%, respectively. Prediction errors varied between antibiotics and were associated with certain phenotypic and genotypic resistance patterns. Conformal prediction reduced errors but increased abstentions; at confidence levels of 90%, 95%, and 97.5%, the model abstained in 9.6%, 14%, and 22% of instances. The method showed promising performance, but its clinical use remains limited and may require diagnostic data beyond susceptibility test results and demographic variables. Importance Antibiotic susceptibility testing is essential for guiding treatment of bacterial infections, but results are often incomplete when initial treatment decisions are made. This study evaluates a novel diagnostic concept: using artificial intelligence to extend the information obtained from partial susceptibility test results, rather than replacing routine susceptibility testing. The method predicts susceptibility to untested antibiotics from patient metadata and existing phenotypic susceptibility results. In this prospective evaluation of clinical Escherichia coli urine isolates, we assessed both prediction performance and uncertainty control by conformal prediction, which allows the method to abstain when predictions are insufficiently reliable. By linking prediction errors to phenotypic and genotypic resistance patterns, the study identified both potential and current limitations of AI-based susceptibility prediction. These findings move AI-based antimicrobial resistance prediction from retrospective model development toward prospective clinical evaluation, a necessary step before implementation.

Infections from antibiotic-resistant bacteria threaten human health globally. Resistance is often caused by mobile antibiotic resistance genes (ARGs) shared horizontally between bacterial genomes. Many ARGs originate from environmental and commensal bacteria and are transferred between divergent bacterial hosts before they reach pathogens. This process remains, however, poorly understood, which complicates the development of countermeasures that reduce the spread of ARGs. In this study, we aimed to systematically analyze the ARGs transferred between the most evolutionarily distant bacteria, defined here based on their phylum. We implemented an algorithm that identified inter-phylum transfers (IPTs) by combining ARG-specific phylogenetic trees with the taxonomy of the bacterial hosts. From the analysis of almost 1 million ARGs identified in >400,000 bacterial genomes, we identified 661 IPTs, which included transfers between all major bacterial phyla. The frequency of IPTs varies substantially between ARG classes and was highest for the aminoglycoside resistance gene AAC(3), while the levels for beta-lactamases were generally lower. ARGs involved in IPTs also differed between phyla, where, for example, tetracycline ARGs were commonly transferred between Firmicutes and Proteobacteria, but rarely between Actinobacteria and Proteobacteria. The results, furthermore, show that conjugative systems are seldom shared between bacterial phyla, suggesting that other mechanisms drive the dissemination of ARGs between divergent hosts. We also show that bacterial genomes involved in IPTs of ARGs are either over- or underrepresented in specific environments. These IPTs were also found to be more recent compared to transfers associated with bacteria isolated from water, soil, and sediment. While macrolide and tetracycline ARGs involved in IPTs almost always were >95% identical between phyla, corresponding β-lactamases showed a median identity of <60%. We conclude that inter-phylum transfer is recurrent, and our results offer new insights into how ARGs are disseminated between evolutionarily distant bacteria. IMPORTANCE:Antibiotic-resistant infections pose a growing threat to global health. This study reveals how genes conferring antibiotic resistance can move between bacteria that belong to different phyla lineages previously thought to be too evolutionarily distant for frequent gene exchange. By analyzing nearly 1 million resistance genes from over 400,000 bacterial genomes, the researchers uncovered hundreds of inter-phylum transfer events, exposing surprising patterns in how different classes of resistance genes spread. The findings highlight that conjugative systems are less common than expected in cross-phyla transfers and suggest that alternative mechanisms may play key roles. This new understanding of how resistance genes leap between vastly different bacterial groups can inform strategies to slow the emergence of drug-resistant infections, aiding in the development of more effective public health interventions.

The dissemination of mobile antibiotic resistance genes (ARGs) via horizontal gene transfer is a significant threat to public health globally. The flow of ARGs into and between pathogens, however, remains poorly understood, limiting our ability to develop strategies for managing the antibiotic resistance crisis. Therefore, we aim to identify genetic and ecological factors that are fundamental for successful horizontal ARG transfer. We used a phylogenetic method to identify instances of horizontal ARG transfer in ~1 million bacterial genomes. This data was then integrated with >20,000 metagenomes representing animal, human, soil, water, and wastewater microbiomes to develop random forest models that can reliably predict horizontal ARG transfer between bacteria. Our results suggest that genetic incompatibility, measured as nucleotide composition dissimilarity, negatively influences the likelihood of transfer of ARGs between evolutionarily divergent bacteria. Conversely, environmental co-occurrence increases the likelihood, especially in humans and wastewater, in which several environment-specific dissemination patterns are observed. This study provides data-driven ways to predict the spread of ARGs and provides insights into the mechanisms governing this evolutionary process.

Streptococcus pyogenes, also known as group A Streptococcus (GAS), is a Gram-positive bacterial pathogen responsible for approximately 500,000 deaths globally per year. Therefore, surveillance of invasive GAS infections is of critical importance in understanding societal transmission. In this study, we propose an optical DNA mapping (ODM)-based assay for strain-level typing of S. pyogenes. Fluorescently labeled DNA molecules of lengths >150 kbp are stretched in nanochannels and imaged to reveal genomic information. A core genome alignment-based typing scheme is then utilized for identification at the strain level. Although our technique lacks the nucleotide-level resolution of whole-genome sequencing, it excels at detecting patterns across long stretches of DNA that remain inaccessible to short-read sequencing methods. Using clinical isolates, we demonstrate that the ODM assay can identify S. pyogenes at both the species level and at even higher taxonomic resolutions with high accuracy. Our typing scheme also shows a high correlation with standard emm subtyping for the analyzed samples. We conclude that ODM has the potential to deliver timely and cost-efficient strain-level bacterial identification.

Antibiotic Resistance (AR) is a critical global health challenge that necessitates the development of cost-effective, efficient, and accurate diagnostic tools. Given the genetic basis of AR, techniques such as Polymerase Chain Reaction (PCR) that target specific resistance genes offer a promising approach for predictive diagnostics using a limited set of key genes. This study introduces GenoARM, a novel framework that integrates reinforcement learning (RL) with transformer-based models to optimize the selection of PCR gene tests and improve AR predictions, leveraging observed metadata for improved accuracy. In our evaluation, we developed several high-performing baselines and compared them using publicly available datasets derived from real-world bacterial samples representing multiple clinically relevant pathogens. The results show that all evaluated methods achieve strong and reliable performance when metadata is not utilized. When metadata is introduced and the number of selected genes increases, GenoARM demonstrates superior performance due to its capacity to approximate rewards for unseen and sparse combinations. Overall, our framework represents a major advancement in optimizing diagnostic tools for AR in clinical settings.

Antibiotic resistance is increasing among bacterial pathogens, posing one of the most severe threats to future public health. A major contributor to the increasing resistance is the dissemination of mobile antibiotic resistance genes (ARGs) among bacterial communities. These genes are ubiquitously present in various environments and are especially diverse in the human gut and wastewater. Despite this, the clinical implications of the prevalence of ARGs in these bacterial communities remain unclear. In this study, we aimed to investigate how the prevalence of ARGs in human gut and wastewater microbiomes reflects their dissemination among important bacterial pathogens. To do this, we estimated the prevalence of >30,000 ARGs, including both well-known (established) and computationally predicted (latent) genes, in >6,000 metagenomic samples. From their prevalence in the human gut and wastewater, we identified four categories of ARGs: co-promoted, human gut (HG)-promoted, wastewater (WW)- promoted, and non-promoted. Our results showed that co-promoted ARGs were by far the most promiscuous, being more frequently found across multiple bacterial phyla, and more often co-localized with broad host range conjugative elements. Co-promoted ARGs were also found to be overrepresented among genes identified in multiple pathogenic species and exhibited an overall higher genetic compatibility with both pathogens and other typical residents of the human gut and wastewater microbiomes. Taken together, our results highlight the link between the promotion of ARGs in the human gut and wastewater microbiomes and their presence in human pathogens, and, thereby, the genes’ potential risk to human health. ### Competing Interest Statement The authors have declared no competing interest. Swedish Research Council, 2018–02835, 2018–05771, 2019–03482, 2022–00945, 2023–01721 Knut and Alice Wallenberg Foundation, KAW 2020.0239 Swedish Foundation for Strategic Research, FFL21-0174

Background Identification of pathogens is crucial to efficiently treat and prevent bacterial infections. However, existing diagnostic techniques are slow or have a too low resolution for well-informed clinical decisions. Methods In this study, we have developed an optical DNA mapping-based method for strain-level bacterial typing and simultaneous plasmid characterisation. For the typing, different taxonomical resolutions were examined and cultivated pure Escherichia coli and Klebsiella pneumoniae samples were used for parameter optimization. Finally, the method was applied to mixed bacterial samples and uncultured urine samples from patients with urinary tract infections. Results We demonstrate that optical DNA mapping of single DNA molecules can identify Escherichia coli and Klebsiella pneumoniae at the strain level directly from patient samples. At a taxonomic resolution corresponding to E. coli sequence type 131 and K. pneumoniae clonal complex 258 forming distinct groups, the average true positive prediction rates are 94% and 89%, respectively. The single-molecule aspect of the method enables us to identify multiple E. coli strains in polymicrobial samples. Furthermore, by targeting plasmid-borne antibiotic resistance genes with Cas9 restriction, we simultaneously identify the strain or subtype and characterize the corresponding plasmids. Conclusion The optical DNA mapping method is accurate and directly applicable to polymicrobial and clinical samples without cultivation. Hence, it has the potential to rapidly provide comprehensive diagnostics information, thereby optimizing early antibiotic treatment and opening up for future precision medicine management.

In 2015, an outbreak caused by OXA-48-producing Enterobacteriaceae affected a neonatal intensive care unit at a Swedish University Hospital. The aim was to explore the transmission of OXA-48-producing strains between infants and the transfer of resistance plasmids between strains during the outbreak. Twenty-four outbreak isolates from ten suspected cases were whole-genome sequenced. A complete assembly was created for the index isolate (Enterobacter cloacae) and used as a mapping reference to detect its plasmids in the remaining isolates (17 Klebsiella pneumoniae, 4 Klebsiella aerogenes, and 2 Escherichia coli). Strain typing was performed using core genome MLST and SNP analysis. As judged from sequencing and clinical epidemiological data, the outbreak involved nine cases (two developed sepsis) and four OXA-48-producing strains: E. cloacae ST1584 (index case), K. pneumoniae ST25 (eight cases), K. aerogenes ST93 (two cases), and E. coli ST453 (2 cases). Two plasmids from the index strain, pEclA2 and pEclA4, carrying blaOXA48 and blaCMY-4, respectively, were traced to all K. pneumoniae ST25 isolates. Klebsiella aerogenes ST93 and E. coli ST453 harboured either only pEclA2, or both pEclA2 and pEclA4. One suspected case harbouring OXA-162-producing K. pneumoniae ST37 could be excluded from the outbreak. Once initiated by an E. cloacae strain, the outbreak was caused by the dissemination of a K. pneumoniae ST25 strain and involved inter-species horizontal transfer of two resistance plasmids, one of which carried blaOXA-48. To our knowledge, this is the first description of an outbreak of OXA-48-producing Enterobacteriaceae in a neonatal setting in northern Europe.

List of new aminoglycoside resistance genes predicted in this study

Background Bacterial communities in humans, animals, and the external environment maintain a large collection of antibiotic resistance genes (ARGs). However, few of these ARGs are well-characterized and thus established in existing resistance gene databases. In contrast, the remaining latent ARGs are typically unknown and overlooked in most sequencing-based studies. Our view of the resistome and its diversity is therefore incomplete, which hampers our ability to assess risk for promotion and spread of yet undiscovered resistance determinants. Results A reference database consisting of both established and latent ARGs (ARGs not present in current resistance gene repositories) was created. By analyzing more than 10,000 metagenomic samples, we showed that latent ARGs were more abundant and diverse than established ARGs in all studied environments, including the human- and animal-associated microbiomes. The pan-resistomes, i.e., all ARGs present in an environment, were heavily dominated by latent ARGs. In comparison, the core-resistome, i.e., ARGs that were commonly encountered, comprised both latent and established ARGs. We identified several latent ARGs shared between environments and/or present in human pathogens. Context analysis of these genes showed that they were located on mobile genetic elements, including conjugative elements. We, furthermore, identified that wastewater microbiomes had a surprisingly large pan- and core-resistome, which makes it a potentially high-risk environment for the mobilization and promotion of latent ARGs. Conclusions Our results show that latent ARGs are ubiquitously present in all environments and constitute a diverse reservoir from which new resistance determinants can be recruited to pathogens. Several latent ARGs already had high mobile potential and were present in human pathogens, suggesting that they may constitute emerging threats to human health. We conclude that the full resistome—including both latent and established ARGs—needs to be considered to properly assess the risks associated with antibiotic selection pressures.

Antibiotic resistance is a growing threat to human health, caused in part by pathogens accumulating antibiotic resistance genes (ARGs) through horizontal gene transfer. New ARGs are typically not recognized until they have become widely disseminated, which limits our ability to reduce their spread. In this study, we use large-scale computational screening of bacterial genomes to identify previously undiscovered mobile ARGs in pathogens. From ~1 million genomes, we predict 1,071,815 genes encoding 34,053 unique aminoglycoside-modifying enzymes (AMEs). These cluster into 7,612 families (<70% amino acid identity) of which 88 are previously described. Fifty new AME families are associated with mobile genetic elements and pathogenic hosts. From these, 24 of 28 experimentally tested AMEs confer resistance to aminoglycoside(s) in Escherichia coli , with 17 providing resistance above clinical breakpoints. This study greatly expands the range of clinically relevant aminoglycoside resistance determinants and demonstrates that computational methods enable early discovery of potentially emerging ARGs.

Macrolides are broad-spectrum antibiotics used to treat a range of infections. Resistance to macrolides is often conferred by mobile resistance genes encoding Erm methyltransferases or Mph phosphotransferases. New erm and mph genes keep being discovered in clinical settings but their origins remain unknown, as is the type of macrolide resistance genes that will appear in the future. In this study, we used optimized hidden Markov models to characterize the macrolide resistome. Over 16 terabases of genomic and metagenomic data, representing a large taxonomic diversity (11 030 species) and diverse environments (1944 metagenomic samples), were searched for the presence of erm and mph genes. From this data, we predicted 28 340 macrolide resistance genes encoding 2892 unique protein sequences, which were clustered into 663 gene families (<70 % amino acid identity), of which 619 (94 %) were previously uncharacterized. This included six new resistance gene families, which were located on mobile genetic elements in pathogens. The function of ten predicted new resistance genes were experimentally validated in Escherichia coli using a growth assay. Among the ten tested genes, seven conferred increased resistance to erythromycin, with five genes additionally conferring increased resistance to azithromycin, showing that our models can be used to predict new functional resistance genes. Our analysis also showed that macrolide resistance genes have diverse origins and have transferred horizontally over large phylogenetic distances into human pathogens. This study expands the known macrolide resistome more than ten-fold, provides insights into its evolution, and demonstrates how computational screening can identify new resistance genes before they become a significant clinical problem.

Nonlinear mixed effects (NLME) modeling is a powerful tool to analyze time-series data from several individual entities in an experiment. In this paper, we give a brief overview of a package for NLME modeling in Wolfram Mathematica entitled NLMEModeling, implementing the first-order conditional estimation method with sensitivity equation-based gradients for parameter estimation. NLMEModeling supports mixed effects modeling of dynamical systems where the underlying dynamics are described by either ordinary or stochastic differential equations combined with observation equations with flexible observation error models. Moreover, NLMEModeling is a user-friendly package with functionality for parameter estimation, model diagnostics (such as goodness-of-fit analysis and visual predictive checks), and model simulation. The package is freely available and provides an extensible add-on to Wolfram Mathematica.

Optical DNA mapping (ODM) is based on fluorescent labeling, stretching and imaging of single DNA molecules to obtain sequence-specific fluorescence profiles, DNA barcodes. These barcodes can be mapped to theoretical counterparts obtained from DNA reference sequences, which in turn allow for DNA identification in complex samples and for detecting structural changes in individual DNA molecules. There are several types of DNA labeling schemes for ODM and for each labeling type one or several types of match scoring methods are used. By combining the information from multiple labeling schemes one can potentially improve mapping confidence; however, combining match scores from different labeling assays has not been implemented yet. In this study, we introduce two theoretical methods for dealing with analysis of DNA molecules with multiple label types. In our first method, we convert the alignment scores, given as output from the different assays, into p-values using carefully crafted null models. We then combine the p-values for different label types using standard methods to obtain a combined match score and an associated combined p-value. In the second method, we use a block bootstrap approach to check for the uniqueness of a match to a database for all barcodes matching with a combined p-value below a predefined threshold. For obtaining experimental dual-labeled DNA barcodes, we introduce a novel assay where we cut plasmid DNA molecules from bacteria with restriction enzymes and the cut sites serve as sequence-specific markers, which together with barcodes obtained using the established competitive binding labeling method, form a dual-labeled barcode. All experimental data in this study originates from this assay, but we point out that our theoretical framework can be used to combine data from all kinds of available optical DNA mapping assays. We test our multiple labeling frameworks on barcodes from two different plasmids and synthetically generated barcodes (combined competitive-binding- and nick-labeling). It is demonstrated that by simultaneously using the information from all label types, we can substantially increase the significance when we match experimental barcodes to a database consisting of theoretical barcodes for all sequenced plasmids.

BACKGROUNDMetallo-β-lactamases (MBLs) are enzymes that use zinc-dependent hydrolysis to confer resistance to almost all available β-lactam antibiotics. They are hypothesized to originate from commensal and environmental bacteria, from where some have mobilized and transferred horizontally to pathogens. The current phylogeny of MBLs, however, is biased as it is founded largely on genes encountered in pathogenic bacteria. This incompleteness is emphasized by recent findings of environmental MBLs with new forms of zinc binding sites and atypical functional profiles.OBJECTIVESTo expand the phylogeny of MBLs to provide a more accurate view of their evolutionary history.METHODSWe searched more than 16 terabases of genomic and metagenomic data for MBLs of the three subclasses B1, B2 and B3 using the validated fARGene method. Predicted genes, together with the previously known ones, were used to infer phylogenetic trees.RESULTSWe identified 2290 unique MBL genes forming 817 gene families, of which 741 were previously uncharacterized. MBLs from subclasses B1 and B3 separated into distinct monophyletic groups, in agreement with their taxonomic and functional properties. We present evidence that clinically associated MBLs were mobilized from Proteobacteria. Additionally, we identified three new variants of the zinc binding sites, indicating that the functional repertoire is broader than previously reported.CONCLUSIONSBased on our results, we recommend that the nomenclature of MBLs is refined into the phylogenetic groups B1.1-B1.5 and B3.1-B3.4 that more accurately describe their molecular and functional characteristics. Our results will also facilitate the annotation of novel MBLs, reflecting their taxonomic organization and evolutionary origin.

Tetracyclines are broad-spectrum antibiotics used to prevent or treat a variety of bacterial infections. Resistance is often mediated through mobile resistance genes, which encode one of the three main mechanisms: active efflux, ribosomal target protection or enzymatic degradation. In the last few decades, a large number of new tetracycline-resistance genes have been discovered in clinical settings. These genes are hypothesized to originate from environmental and commensal bacteria, but the diversity of tetracycline-resistance determinants that have not yet been mobilized into pathogens is unknown. In this study, we aimed to characterize the potential tetracycline resistome by screening genomic and metagenomic data for novel resistance genes. By using probabilistic models, we predicted 1254 unique putative tetracycline resistance genes, representing 195 gene families (<70% amino acid sequence identity), whereof 164 families had not been described previously. Out of 17 predicted genes selected for experimental verification, 7 induced a resistance phenotype in an Escherichia coli host. Several of the predicted genes were located on mobile genetic elements or in regions that indicated mobility, suggesting that they easily can be shared between bacteria. Furthermore. phylogenetic analysis indicated several events of horizontal gene transfer between bacterial phyla. Our results also suggested that acquired efflux pumps originate from proteobacterial species, while ribosomal protection genes have been mobilized from Firmicutes and Actinobacteria. This study significantly expands the knowledge of known and putatively novel tetracycline resistance genes, their mobility and evolutionary history. The study also provides insights into the unknown resistome and genes that may be encountered in clinical settings in the future.

Nonlinear mixed effects modeling is a powerful tool when analyzing data from several entities in an experiment. In this paper, we present NLMEModeling, a package for mixed effects modeling in Wolfram Mathematica. NLMEModeling supports mixed effects modeling of dynamical systems where the underlying dynamics are described by either ordinary or stochastic differential equations combined with a flexible observation error model. Moreover, NLMEModeling is a user-friendly package with functionality for model validation, visual predictive checks and simulation capabilities. The package is freely available and provides a flexible add-on to Wolfram Mathematica.

A variety of pathogenic bacteria can infect humans, and rapid species identification is crucial for the correct treatment. However, the identification process can often be time-consuming and depend on the cultivation of the bacterial pathogen(s). Here, we present a stand-alone, enzyme-free, optical DNA mapping assay capable of species identification by matching the intensity profiles of large DNA molecules to a database of fully assembled bacterial genomes (>10 000). The assay includes a new data analysis strategy as well as a general DNA extraction protocol for both Gram-negative and Gram-positive bacteria. We demonstrate that the assay is capable of identifying bacteria directly from uncultured clinical urine samples, as well as in mixtures, with the potential to be discriminative even at the subspecies level. We foresee that the assay has applications both within research laboratories and in clinical settings, where the time-consuming step of cultivation can be minimized or even completely avoided.