Phenetic Comparison of Prokaryotic Genomes Using k-mers

Classification of Neisseria meningitidis genomes with a bag-of-words approach and machine learning

Whole genome sequencing of bacteria is important to enable strain classification. Using entire genomes as an input to machine learning (ML) models would allow rapid classification of strains while using information from multiple genetic elements. We developed a "bag-of-words" approach to encode, using SentencePiece or k-mer tokenization, entire bacterial genomes and analyze these with ML. Initial model selection identified SentencePiece with 8,000 and 32,000 words as the best approach for genome tokenization. We then classified in Neisseria meningitidis genomes the capsule B group genotype with 99.6% accuracy and the multifactor invasive phenotype with 90.2% accuracy, in an independent test set. Subsequently, in silico knockouts of 2,808 genes confirmed that the ML model predictions aligned with our current understanding of the underlying biology. To our knowledge, this is the first ML method using entire bacterial genomes to classify strains and identify genes considered relevant by the classifier.

Gene-based microbiome representation enhances host phenotype classification

With the concomitant advances in both the microbiome and machine learning fields, the gut microbiome has become of great interest for the potential discovery of biomarkers to be used in the classification of the host health status. Shotgun metagenomics data derived from the human microbiome is composed of a high-dimensional set of microbial features. The use of such complex data for the modeling of host-microbiome interactions remains a challenge as retaining de novo content yields a highly granular set of microbial features. In this study, we compared the prediction performances of machine learning approaches according to different types of data representations derived from shotgun metagenomics. These representations include commonly used taxonomic and functional profiles and the more granular gene cluster approach. For the five case-control datasets used in this study (Type 2 diabetes, obesity, liver cirrhosis, colorectal cancer, and inflammatory bowel disease), gene-based approaches, whether used alone or in combination with reference-based data types, allowed improved or similar classification performances as the taxonomic and functional profiles. In addition, we show that using subsets of gene families from specific functional categories of genes highlight the importance of these functions on the host phenotype. This study demonstrates that both reference-free microbiome representations and curated metagenomic annotations can provide relevant representations for machine learning based on metagenomic data. IMPORTANCE Data representation is an essential part of machine learning performance when using metagenomic data. In this work, we show that different microbiome representations provide varied host phenotype classification performance depending on the dataset. In classification tasks, untargeted microbiome gene content can provide similar or improved classification compared to taxonomical profiling. Feature selection based on biological function also improves classification performance for some pathologies. Function-based feature selection combined with interpretable machine learning algorithms can generate new hypotheses that can potentially be assayed mechanistically. This work thus proposes new approaches to represent microbiome data for machine learning that can potentiate the findings associated with metagenomic data.

Gut metagenome profile of the Nunavik Inuit youth is distinct from industrial and non-industrial counterparts

Comparative metagenomics studies have highlighted differences in microbiome community structure among human populations over diverse lifestyles and environments. With their unique environmental and historical backgrounds, Nunavik Inuit have a distinctive gut microbiome with undocumented health-related implications. Using shotgun metagenomics, we explored the taxonomic and functional structure of the gut microbiome from 275 Nunavik Inuit ranging from 16 to 30-year-old. Whole-metagenome analyses revealed that Nunavik Inuit youths have a more diverse microbiome than their non-industrialized and industrialized counterparts. A comparison of k-mer content illustrated the uniqueness of the Nunavik gut microbiome. Short-chain fatty acids producing species, and carbohydrates degradation pathways dominated Inuit metagenomes. We identified a taxonomic and functional signature unique to the Nunavik gut microbiome contrasting with other populations using a random forest classifier. Here, we show that the Nunavik Inuit gut microbiome exhibits high diversity and a distinct community structure.

Antimicrobial Resistance in the Environment: Towards Elucidating the Roles of Bioaerosols in Transmission and Detection of Antibacterial Resistance Genes

Antimicrobial resistance (AMR) is continuing to grow across the world. Though often thought of as a mostly public health issue, AMR is also a major agricultural and environmental problem. As such, many researchers refer to it as the preeminent One Health issue. Aerial transport of antimicrobial-resistant bacteria via bioaerosols is still poorly understood. Recent work has highlighted the presence of antibiotic resistance genes in bioaerosols. Emissions of AMR bacteria and genes have been detected from various sources, including wastewater treatment plants, hospitals, and agricultural practices; however, their impacts on the broader environment are poorly understood. Contextualizing the roles of bioaerosols in the dissemination of AMR necessitates a multidisciplinary approach. Environmental factors, industrial and medical practices, as well as ecological principles influence the aerial dissemination of resistant bacteria. This article introduces an ongoing project assessing the presence and fate of AMR in bioaerosols across Canada. Its various sub-studies include the assessment of the emissions of antibiotic resistance genes from many agricultural practices, their long-distance transport, new integrative methods of assessment, and the creation of dissemination models over short and long distances. Results from sub-studies are beginning to be published. Consequently, this paper explains the background behind the development of the various sub-studies and highlight their shared aspects.

Genomic diversity and CRISPR-Cas systems in the cyanobacterium Nostoc in the High Arctic

Nostoc (Nostocales, Cyanobacteria) has a global distribution in the Polar Regions. However, the genomic diversity of Nostoc is little known and there are no genomes available for polar Nostoc. Here we carried out the first genomic analysis of the Nostoc commune morphotype with a recent sample from the High Arctic and a herbarium specimen collected during the British Arctic Expedition (1875-76). Comparisons of the polar genomes with 26 present-day non-polar members of the Nostocales family highlighted that there are pronounced genetic variations among Nostoc strains and species. Osmoprotection and other stress genes were found in all Nostoc strains, but the two Arctic strains had markedly higher numbers of biosynthetic gene clusters for uncharacterised non-ribosomal peptide synthetases, suggesting a high diversity of secondary metabolites. Since viral-host interactions contribute to microbial diversity, we analysed the CRISPR-Cas systems in the Arctic and two temperate Nostoc species. There were a large number of unique repeat-spacer arrays in each genome, indicating diverse histories of viral attack. All Nostoc strains had a subtype I-D system, but the polar specimens also showed evidence of a subtype I-B system that has not been previously reported in cyanobacteria, suggesting diverse cyanobacteria-virus interactions in the Arctic.

Exploring short k-mer profiles in cells and mobile elements from Archaea highlights the major influence of both the ecological niche and evolutionary history

BACKGROUND

K-mer-based methods have greatly advanced in recent years, largely driven by the realization of their biological significance and by the advent of next-generation sequencing. Their speed and their independence from the annotation process are major advantages. Their utility in the study of the mobilome has recently emerged and they seem a priori adapted to the patchy gene distribution and the lack of universal marker genes of viruses and plasmids. To provide a framework for the interpretation of results from k-mer based methods applied to archaea or their mobilome, we analyzed the 5-mer DNA profiles of close to 600 archaeal cells, viruses and plasmids. Archaea is one of the three domains of life. Archaea seem enriched in extremophiles and are associated with a high diversity of viral and plasmid families, many of which are specific to this domain. We explored the dataset structure by multivariate and statistical analyses, seeking to identify the underlying factors.

RESULTS

For cells, the 5-mer profiles were inconsistent with the phylogeny of archaea. At a finer taxonomic level, the influence of the taxonomy and the environmental constraints on 5-mer profiles was very strong. These two factors were interdependent to a significant extent, and the respective weights of their contributions varied according to the clade. A convergent adaptation was observed for the class Halobacteria, for which a strong 5-mer signature was identified. For mobile elements, coevolution with the host had a clear influence on their 5-mer profile. This enabled us to identify one previously known and one new case of recent host transfer based on the atypical composition of the mobile elements involved. Beyond the effect of coevolution, extrachromosomal elements strikingly retain the specific imprint of their own viral or plasmid taxonomic family in their 5-mer profile.

CONCLUSION

This specific imprint confirms that the evolution of extrachromosomal elements is driven by multiple parameters and is not restricted to host adaptation. In addition, we detected only recent host transfer events, suggesting the fast evolution of short k-mer profiles. This calls for caution when using k-mers for host prediction, metagenomic binning or phylogenetic reconstruction.

BACTOME-a reference database to explore the sequence- and gene expression-variation landscape of Pseudomonas aeruginosa clinical isolates

Extensive use of next-generation sequencing (NGS) for pathogen profiling has the potential to transform our understanding of how genomic plasticity contributes to phenotypic versatility. However, the storage of large amounts of NGS data and visualization tools need to evolve to offer the scientific community fast and convenient access to these data. We introduce BACTOME as a database system that links aligned DNA- and RNA-sequencing reads of clinical Pseudomonas aeruginosa isolates with clinically relevant pathogen phenotypes. The database allows data extraction for any single isolate, gene or phenotype as well as data filtering and phenotypic grouping for specific research questions. With the integration of statistical tools we illustrate the usefulness of a relational database structure for the identification of phenotype-genotype correlations as an essential part of the discovery pipeline in genomic research. Furthermore, the database provides a compilation of DNA sequences and gene expression values of a plethora of clinical isolates to give a consensus DNA sequence and consensus gene expression signature. Deviations from the consensus thereby describe the genomic landscape and the transcriptional plasticity of the species P. aeruginosa. The database is available at https://bactome.helmholtz-hzi.de.

Phenotypic and Genetic Characterization of the Cheese Ripening Yeast Geotrichum candidum

The yeast Geotrichum candidum (teleomorph Galactomyces candidus) is inoculated onto mold- and smear-ripened cheeses and plays several roles during cheese ripening. Its ability to metabolize proteins, lipids, and organic acids enables its growth on the cheese surface and promotes the development of organoleptic properties. Recent multilocus sequence typing (MLST) and phylogenetic analyses of G. candidum isolates revealed substantial genetic diversity, which may explain its strain-dependant technological capabilities. Here, we aimed to shed light on the phenotypic and genetic diversity among eight G. candidum and three Galactomyces spp. strains of environmental and dairy origin. Phenotypic tests such as carbon assimilation profiles, the ability to grow at 35°C and morphological traits on agar plates allowed us to discriminate G. candidum from Galactomyces spp. The genomes of these isolates were sequenced and assembled; whole genome comparison clustered the G. candidum strains into three subgroups and provided a reliable reference for MLST scheme optimization. Using the whole genome sequence as a reference, we optimized an MLST scheme using six loci that were proposed in two previous MLST schemes. This new MLST scheme allowed us to identify 15 sequence types (STs) out of 41 strains and revealed three major complexes named GeoA, GeoB, and GeoC. The population structure of these 41 strains was evaluated with STRUCTURE and a NeighborNet analysis of the combined six loci, which revealed recombination events between and within the complexes. These results hint that the allele variation conferring the different STs arose from recombination events. Recombination occurred for the six housekeeping genes studied, but most likely occurred throughout the genome. These recombination events may have induced an adaptive divergence between the wild strains and the cheesemaking strains, as observed for other cheese ripening fungi. Further comparative genomic studies are needed to confirm this phenomenon in G. candidum. In conclusion, the draft assembly of 11 G. candidum/Galactomyces spp. genomes allowed us to optimize a genotyping MLST scheme and, combined with the assessment of their ability to grow under different conditions, provides a reliable tool to cluster and eventually improves the selection of G. candidum strains.

Evaluation of Machine Learning Models for Predicting Antimicrobial Resistance of Actinobacillus pleuropneumoniae From Whole Genome Sequences

Antimicrobial resistance (AMR) is becoming a huge problem in countries all over the world, and new approaches to identifying strains resistant or susceptible to certain antibiotics are essential in fighting against antibiotic-resistant pathogens. Genotype-based machine learning methods showed great promise as a diagnostic tool, due to the increasing availability of genomic datasets and AST phenotypes. In this article, Support Vector Machine (SVM) and Set Covering Machine (SCM) models were used to learn and predict the resistance of the five drugs (Tetracycline, Ampicillin, Sulfisoxazole, Trimethoprim, and Enrofloxacin). The SVM model used the number of co-occurring k-mers between the genome of the isolates and the reference genes to learn and predict the phenotypes of the bacteria to a specific antimicrobial, while the SCM model uses a greedy approach to construct conjunction or disjunction of Boolean functions to find the most concise set of k-mers that allows for accurate prediction of the phenotype. Five-fold cross-validation was performed on the training set of the SVM and SCM model to select the best hyperparameter values to avoid model overfitting. The training accuracy (mean cross-validation score) and the testing accuracy of SVM and SCM models of five drugs were above 90% regardless of the resistant mechanism of which were acquired resistant or point mutation in the chromosome. The results of correlation between the phenotype and the model predictions of the five drugs indicated that both SVM and SCM models could significantly classify the resistant isolates from the sensitive isolates of the bacteria (p < 0.01), and would be used as potential tools in antimicrobial resistance surveillance and clinical diagnosis in veterinary medicine.

Extended sequencing of vaccine and wild-type capripoxvirus isolates provides insights into genes modulating virulence and host range

The genus Capripoxvirus in the subfamily Chordopoxvirinae, family Poxviridae, comprises sheeppox virus (SPPV), goatpox virus (GTPV) and lumpy skin disease virus (LSDV), which cause the eponymous diseases across parts of Africa, the Middle East and Asia. These diseases cause significant economic losses and can have a devastating impact on the livelihoods and food security of small farm holders. So far, only live classically attenuated SPPV, GTPV and LSDV vaccines are commercially available and the history, safety and efficacy of many have not been well established. Here, we report 13 new capripoxvirus genome sequences, including the hairpin telomeres, from both pathogenic field isolates and vaccine strains. We have also updated the genome annotations to incorporate recent advances in our understanding of poxvirus biology. These new genomes and genes grouped phenetically with other previously sequenced capripoxvirus strains, and these new alignments collectively identified several recurring alterations in genes thought to modulate virulence and host range. In particular, some of the many large capripoxvirus ankyrin and kelch-like proteins are commonly mutated in vaccine strains, while the variola virus B22R-like gene homolog has also been disrupted in many vaccine isolates. Among these vaccine isolates, frameshift mutations are especially common and clearly present a risk of reversion to wild type in vaccines bearing these mutations. A consistent pattern of gene inactivation from LSDV to GTPV and then SPPV is also observed, much like the pattern of gene loss in orthopoxviruses, but, rather surprisingly, the overall genome size of ~150 kbp remains relatively constant. These data provide new insights into the evolution of capripoxviruses and the determinants of pathogenicity and host range. They will find application in the development of new vaccines with better safety, efficacy and trade profiles.

Interpretable genotype-to-phenotype classifiers with performance guarantees

Understanding the relationship between the genome of a cell and its phenotype is a central problem in precision medicine. Nonetheless, genotype-to-phenotype prediction comes with great challenges for machine learning algorithms that limit their use in this setting. The high dimensionality of the data tends to hinder generalization and challenges the scalability of most learning algorithms. Additionally, most algorithms produce models that are complex and difficult to interpret. We alleviate these limitations by proposing strong performance guarantees, based on sample compression theory, for rule-based learning algorithms that produce highly interpretable models. We show that these guarantees can be leveraged to accelerate learning and improve model interpretability. Our approach is validated through an application to the genomic prediction of antimicrobial resistance, an important public health concern. Highly accurate models were obtained for 12 species and 56 antibiotics, and their interpretation revealed known resistance mechanisms, as well as some potentially new ones. An open-source disk-based implementation that is both memory and computationally efficient is provided with this work. The implementation is turnkey, requires no prior knowledge of machine learning, and is complemented by comprehensive tutorials.