Optimized model architectures for deep learning on genomic data.

The success of deep learning in various applications depends on task-specific architecture design choices, including the types, hyperparameters, and number of layers. In computational biology, there is no consensus on the optimal architecture design, and decisions are often made using insights from more well-established fields such as computer vision. These may not consider the domain-specific characteristics of genome sequences, potentially limiting performance. Here, we present GenomeNet-Architect, a neural architecture design framework that automatically optimizes deep learning models for genome sequence data. It optimizes the overall layout of the architecture, with a search space specifically designed for genomics. Additionally, it optimizes hyperparameters of individual layers and the model training procedure. On a viral classification task, GenomeNet-Architect reduced the read-level misclassification rate by 19%, with 67% faster inference and 83% fewer parameters, and achieved similar contig-level accuracy with ~100 times fewer parameters compared to the best-performing deep learning baselines.

A host-microbiota interactome reveals extensive transkingdom connectivity.

The myriad microorganisms that live in close association with humans have diverse effects on physiology, yet the molecular bases for these impacts remain mostly unknown1-3. Classical pathogens often invade host tissues and modulate immune responses through interactions with human extracellular and secreted proteins (the 'exoproteome'). Commensal microorganisms may also facilitate niche colonization and shape host biology by engaging host exoproteins; however, direct exoproteome-microbiota interactions remain largely unexplored. Here we developed and validated a novel technology, BASEHIT, that enables proteome-scale assessment of human exoproteome-microbiome interactions. Using BASEHIT, we interrogated more than 1.7 million potential interactions between 519 human-associated bacterial strains from diverse phylogenies and tissues of origin and 3,324 human exoproteins. The resulting interactome revealed an extensive network of transkingdom connectivity consisting of thousands of previously undescribed host-microorganism interactions involving 383 strains and 651 host proteins. Specific binding patterns within this network implied underlying biological logic; for example, conspecific strains exhibited shared exoprotein-binding patterns, and individual tissue isolates uniquely bound tissue-specific exoproteins. Furthermore, we observed dozens of unique and often strain-specific interactions with potential roles in niche colonization, tissue remodelling and immunomodulation, and found that strains with differing host interaction profiles had divergent interactions with host cells in vitro and effects on the host immune system in vivo. Overall, these studies expose a previously unexplored landscape of molecular-level host-microbiota interactions that may underlie causal effects of indigenous microorganisms on human health and disease.

Discovering and exploring the hidden diversity of human gut viruses using highly enriched virome samples.

Viruses are an abundant and crucial component of the human microbiome, but accurately discovering them via metagenomics is still challenging. Currently, the available viral reference genomes poorly represent the diversity in microbiome samples, and expanding such a set of viral references is difficult. As a result, many viruses are still undetectable through metagenomics even when considering the power of de novo metagenomic assembly and binning, as viruses lack universal markers. Here, we describe a novel approach to catalog new viral members of the human gut microbiome and show how the resulting resource improves metagenomic analyses. We retrieved >3,000 viral-like particles (VLP) enriched metagenomic samples (viromes), evaluated the efficiency of the enrichment in each sample to leverage the viromes of highest purity, and applied multiple analysis steps involving assembly and comparison with hundreds of thousands of metagenome-assembled genomes to discover new viral genomes. We reported over 162,000 viral sequences passing quality control from thousands of gut metagenomes and viromes. The great majority of the retrieved viral sequences (~94.4%) were of unknown origin, most had a CRISPR spacer matching host bacteria, and four of them could be detected in >50% of a set of 18,756 gut metagenomes we surveyed. We included the obtained collection of sequences in a new MetaPhlAn 4.1 release, which can quantify reads within a metagenome matching the known and newly uncovered viral diversity. Additionally, we released the viral database for further virome and metagenomic studies of the human microbiome.

Integrated annotation prioritizes metabolites with bioactivity in inflammatory bowel disease.

Microbial biochemistry is central to the pathophysiology of inflammatory bowel diseases (IBD). Improved knowledge of microbial metabolites and their immunomodulatory roles is thus necessary for diagnosis and management. Here, we systematically analyzed the chemical, ecological, and epidemiological properties of ~82k metabolic features in 546 Integrative Human Microbiome Project (iHMP/HMP2) metabolomes, using a newly developed methodology for bioactive compound prioritization from microbial communities. This suggested >1000 metabolic features as potentially bioactive in IBD and associated ~43% of prevalent, unannotated features with at least one well-characterized metabolite, thereby providing initial information for further characterization of a significant portion of the fecal metabolome. Prioritized features included known IBD-linked chemical families such as bile acids and short-chain fatty acids, and less-explored bilirubin, polyamine, and vitamin derivatives, and other microbial products. One of these, nicotinamide riboside, reduced colitis scores in DSS-treated mice. The method, MACARRoN, is generalizable with the potential to improve microbial community characterization and provide therapeutic candidates.

BugSigDB captures patterns of differential abundance across a broad range of host-associated microbial signatures.

The literature of human and other host-associated microbiome studies is expanding rapidly, but systematic comparisons among published results of host-associated microbiome signatures of differential abundance remain difficult. We present BugSigDB, a community-editable database of manually curated microbial signatures from published differential abundance studies accompanied by information on study geography, health outcomes, host body site and experimental, epidemiological and statistical methods using controlled vocabulary. The initial release of the database contains >2,500 manually curated signatures from >600 published studies on three host species, enabling high-throughput analysis of signature similarity, taxon enrichment, co-occurrence and coexclusion and consensus signatures. These data allow assessment of microbiome differential abundance within and across experimental conditions, environments or body sites. Database-wide analysis reveals experimental conditions with the highest level of consistency in signatures reported by independent studies and identifies commonalities among disease-associated signatures, including frequent introgression of oral pathobionts into the gut.

Profiling novel lateral gene transfer events in the human microbiome.

Lateral gene transfer (LGT) is an important mechanism for genome diversification in microbial populations, including the human microbiome. While prior work has surveyed LGT events in human-associated microbial isolate genomes, the scope and dynamics of novel LGT events arising in personal microbiomes are not well understood, as there are no widely adopted computational methods to detect, quantify, and characterize LGT from complex microbial communities. We addressed this by developing, benchmarking, and experimentally validating a computational method (WAAFLE) to profile novel LGT events from assembled metagenomes. Applying WAAFLE to >2K human metagenomes from diverse body sites, we identified >100K putative high-confidence but previously uncharacterized LGT events (~2 per assembled microbial genome-equivalent). These events were enriched for mobile elements (as expected), as well as restriction-modification and transport functions typically associated with the destruction of foreign DNA. LGT frequency was quantifiably influenced by biogeography, the phylogenetic similarity of the involved taxa, and the ecological abundance of the donor taxon. These forces manifest as LGT networks in which hub species abundant in a community type donate unequally with their close phylogenetic neighbors. Our findings suggest that LGT may be a more ubiquitous process in the human microbiome than previously described. The open-source WAAFLE implementation, documentation, and data from this work are available at http://huttenhower.sph.harvard.edu/waafle.

Pulsed antibiotic treatments of gnotobiotic mice manifest in complex bacterial community dynamics and resistance effects.

Bacteria can evolve to withstand a wide range of antibiotics (ABs) by using various resistance mechanisms. How ABs affect the ecology of the gut microbiome is still poorly understood. We investigated strain-specific responses and evolution during repeated AB perturbations by three clinically relevant ABs, using gnotobiotic mice colonized with a synthetic bacterial community (oligo-mouse-microbiota). Over 80 days, we observed resilience effects at the strain and community levels, and we found that they were correlated with modulations of the estimated growth rate and levels of prophage induction as determined from metagenomics data. Moreover, we tracked mutational changes in the bacterial populations, and this uncovered clonal expansion and contraction of haplotypes and selection of putative AB resistance-conferring SNPs. We functionally verified these mutations via reisolation of clones with increased minimum inhibitory concentration (MIC) of ciprofloxacin and tetracycline from evolved communities. This demonstrates that host-associated microbial communities employ various mechanisms to respond to selective pressures that maintain community stability.

RNA-based amplicon sequencing is ineffective in measuring metabolic activity in environmental microbial communities.

BACKGROUND: Characterization of microbial activity is essential to the understanding of the basic biology of microbial communities, as the function of a microbiome is defined by its biochemically active ("viable") community members. Current sequence-based technologies can rarely differentiate microbial activity, due to their inability to distinguish live and dead sourced DNA. As a result, our understanding of microbial community structures and the potential mechanisms of transmission between humans and our surrounding environments remains incomplete. As a potential solution, 16S rRNA transcript-based amplicon sequencing (16S-RNA-seq) has been proposed as a reliable methodology to characterize the active components of a microbiome, but its efficacy has not been evaluated systematically. Here, we present our work to benchmark RNA-based amplicon sequencing for activity assessment in synthetic and environmentally sourced microbial communities. RESULTS: In synthetic mixtures of living and heat-killed Escherichia coli and Streptococcus sanguinis, 16S-RNA-seq successfully reconstructed the active compositions of the communities. However, in the realistic environmental samples, no significant compositional differences were observed in RNA ("actively transcribed - active") vs. DNA ("whole" communities) spiked with E. coli controls, suggesting that this methodology is not appropriate for activity assessment in complex communities. The results were slightly different when validated in environmental samples of similar origins (i.e., from Boston subway systems), where samples were differentiated both by environment type as well as by library type, though compositional dissimilarities between DNA and RNA samples remained low (Bray-Curtis distance median: 0.34-0.49). To improve the interpretation of 16S-RNA-seq results, we compared our results with previous studies and found that 16S-RNA-seq suggests taxon-wise viability trends (i.e., specific taxa are universally more or less likely to be viable compared to others) in samples of similar origins. CONCLUSIONS: This study provides a comprehensive evaluation of 16S-RNA-seq for viability assessment in synthetic and complex microbial communities. The results found that while 16S-RNA-seq was able to semi-quantify microbial viability in relatively simple communities, it only suggests a taxon-dependent "relative" viability in realistic communities.  Video Abstract.

Gut microbial metabolism of 5-ASA diminishes its clinical efficacy in inflammatory bowel disease.

For decades, variability in clinical efficacy of the widely used inflammatory bowel disease (IBD) drug 5-aminosalicylic acid (5-ASA) has been attributed, in part, to its acetylation and inactivation by gut microbes. Identification of the responsible microbes and enzyme(s), however, has proved elusive. To uncover the source of this metabolism, we developed a multi-omics workflow combining gut microbiome metagenomics, metatranscriptomics and metabolomics from the longitudinal IBDMDB cohort of 132 controls and patients with IBD. This associated 12 previously uncharacterized microbial acetyltransferases with 5-ASA inactivation, belonging to two protein superfamilies: thiolases and acyl-CoA N-acyltransferases. In vitro characterization of representatives from both families confirmed the ability of these enzymes to acetylate 5-ASA. A cross-sectional analysis within the discovery cohort and subsequent prospective validation within the independent SPARC IBD cohort (n = 208) found three of these microbial thiolases and one acyl-CoA N-acyltransferase to be epidemiologically associated with an increased risk of treatment failure among 5-ASA users. Together, these data address a longstanding challenge in IBD management, outline a method for the discovery of previously uncharacterized gut microbial activities and advance the possibility of microbiome-based personalized medicine.

Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4.

Metagenomic assembly enables new organism discovery from microbial communities, but it can only capture few abundant organisms from most metagenomes. Here we present MetaPhlAn 4, which integrates information from metagenome assemblies and microbial isolate genomes for more comprehensive metagenomic taxonomic profiling. From a curated collection of 1.01 M prokaryotic reference and metagenome-assembled genomes, we define unique marker genes for 26,970 species-level genome bins, 4,992 of them taxonomically unidentified at the species level. MetaPhlAn 4 explains ~20% more reads in most international human gut microbiomes and >40% in less-characterized environments such as the rumen microbiome and proves more accurate than available alternatives on synthetic evaluations while also reliably quantifying organisms with no cultured isolates. Application of the method to >24,500 metagenomes highlights previously undetected species to be strong biomarkers for host conditions and lifestyles in human and mouse microbiomes and shows that even previously uncharacterized species can be genetically profiled at the resolution of single microbial strains.

Microbiome epidemiology and association studies in human health.

Studies of the human microbiome share both technical and conceptual similarities with genome-wide association studies and genetic epidemiology. However, the microbiome has many features that differ from genomes, such as its temporal and spatial variability, highly distinct genetic architecture and person-to-person variation. Moreover, there are various potential mechanisms by which distinct aspects of the human microbiome can relate to health outcomes. Recent advances, including next-generation sequencing and the proliferation of multi-omic data types, have enabled the exploration of the mechanisms that connect microbial communities to human health. Here, we review the ways in which features of the microbiome at various body sites can influence health outcomes, and we describe emerging opportunities and future directions for advanced microbiome epidemiology.

Population structure discovery in meta-analyzed microbial communities and inflammatory bowel disease using MMUPHin.

Microbiome studies of inflammatory bowel diseases (IBD) have achieved a scale for meta-analysis of dysbioses among populations. To enable microbial community meta-analyses generally, we develop MMUPHin for normalization, statistical meta-analysis, and population structure discovery using microbial taxonomic and functional profiles. Applying it to ten IBD cohorts, we identify consistent associations, including novel taxa such as Acinetobacter and Turicibacter, and additional exposure and interaction effects. A single gradient of dysbiosis severity is favored over discrete types to summarize IBD microbiome population structure. These results provide a benchmark for characterization of IBD and a framework for meta-analysis of any microbial communities.

Progress and opportunities in microbial community metabolomics.

The metabolome lies at the interface of host-microbiome crosstalk. Previous work has established links between chemically diverse microbial metabolites and a myriad of host physiological processes and diseases. Coupled with scalable and cost-effective technologies, metabolomics is thus gaining popularity as a tool for characterization of microbial communities, particularly when combined with metagenomics as a window into microbiome function. A systematic interrogation of microbial community metabolomes can uncover key microbial compounds, metabolic capabilities of the microbiome, and also provide critical mechanistic insights into microbiome-linked host phenotypes. In this review, we discuss methods and accompanying resources that have been developed for these purposes. The accomplishments of these methods demonstrate that metabolomes can be used to functionally characterize microbial communities, and that microbial properties can be used to identify and investigate chemical compounds.

A biosynthetic pathway for the selective sulfonation of steroidal metabolites by human gut bacteria.

Members of the human gut microbiome enzymatically process many bioactive molecules in the gastrointestinal tract. Most gut bacterial modifications characterized so far are hydrolytic or reductive in nature. Here we report that abundant human gut bacteria from the phylum Bacteroidetes perform conjugative modifications by selectively sulfonating steroidal metabolites. While sulfonation is a ubiquitous biochemical modification, this activity has not yet been characterized in gut microbes. Using genetic and biochemical approaches, we identify a widespread biosynthetic gene cluster that encodes both a sulfotransferase (BtSULT, BT0416) and enzymes that synthesize the sulfonate donor adenosine 3'-phosphate-5'-phosphosulfate (PAPS), including an APS kinase (CysC, BT0413) and an ATP sulfurylase (CysD and CysN, BT0414-BT0415). BtSULT selectively sulfonates steroidal metabolites with a flat A/B ring fusion, including cholesterol. Germ-free mice monocolonized with Bacteroides thetaiotaomicron ΔBT0416 exhibited reduced gastrointestinal levels of cholesterol sulfate (Ch-S) compared with wild-type B. thetaiotaomicron-colonized mice. The presence of BtSULT and BtSULT homologues in bacteria inhibited leucocyte migration in vitro and in vivo, and abundances of cluster genes were significantly reduced in patients with inflammatory bowel disease. Together, these data provide a mechanism by which gut bacteria sulfonate steroidal metabolites and suggest that these compounds can modulate immune cell trafficking in the host.

High-sensitivity pattern discovery in large, paired multiomic datasets.

MOTIVATION: Modern biological screens yield enormous numbers of measurements, and identifying and interpreting statistically significant associations among features are essential. In experiments featuring multiple high-dimensional datasets collected from the same set of samples, it is useful to identify groups of associated features between the datasets in a way that provides high statistical power and false discovery rate (FDR) control. RESULTS: Here, we present a novel hierarchical framework, HAllA (Hierarchical All-against-All association testing), for structured association discovery between paired high-dimensional datasets. HAllA efficiently integrates hierarchical hypothesis testing with FDR correction to reveal significant linear and non-linear block-wise relationships among continuous and/or categorical data. We optimized and evaluated HAllA using heterogeneous synthetic datasets of known association structure, where HAllA outperformed all-against-all and other block-testing approaches across a range of common similarity measures. We then applied HAllA to a series of real-world multiomics datasets, revealing new associations between gene expression and host immune activity, the microbiome and host transcriptome, metabolomic profiling and human health phenotypes. AVAILABILITY AND IMPLEMENTATION: An open-source implementation of HAllA is freely available at http://huttenhower.sph.harvard.edu/halla along with documentation, demo datasets and a user group. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Discovery of bioactive microbial gene products in inflammatory bowel disease.

Microbial communities and their associated bioactive compounds1-3 are often disrupted in conditions such as the inflammatory bowel diseases (IBD)4. However, even in well-characterized environments (for example, the human gastrointestinal tract), more than one-third of microbial proteins are uncharacterized and often expected to be bioactive5-7. Here we systematically identified more than 340,000 protein families as potentially bioactive with respect to gut inflammation during IBD, about half of which have not to our knowledge been functionally characterized previously on the basis of homology or experiment. To validate prioritized microbial proteins, we used a combination of metagenomics, metatranscriptomics and metaproteomics to provide evidence of bioactivity for a subset of proteins that are involved in host and microbial cell-cell communication in the microbiome; for example, proteins associated with adherence or invasion processes, and extracellular von Willebrand-like factors. Predictions from high-throughput data were validated using targeted experiments that revealed the differential immunogenicity of prioritized Enterobacteriaceae pilins and the contribution of homologues of von Willebrand factors to the formation of Bacteroides biofilms in a manner dependent on mucin levels. This methodology, which we term MetaWIBELE (workflow to identify novel bioactive elements in the microbiome), is generalizable to other environmental communities and human phenotypes. The prioritized results provide thousands of candidate microbial proteins that are likely to interact with the host immune system in IBD, thus expanding our understanding of potentially bioactive gene products in chronic disease states and offering a rational compendium of possible therapeutic compounds and targets.

Human gut bacteria produce ΤΗ17-modulating bile acid metabolites.

The microbiota modulates gut immune homeostasis. Bacteria influence the development and function of host immune cells, including T helper cells expressing interleukin-17A (TH17 cells). We previously reported that the bile acid metabolite 3-oxolithocholic acid (3-oxoLCA) inhibits TH17 cell differentiation1. Although it was suggested that gut-residing bacteria produce 3-oxoLCA, the identity of such bacteria was unknown, and it was unclear whether 3-oxoLCA and other immunomodulatory bile acids are associated with inflammatory pathologies in humans. Here we identify human gut bacteria and corresponding enzymes that convert the secondary bile acid lithocholic acid into 3-oxoLCA as well as the abundant gut metabolite isolithocholic acid (isoLCA). Similar to 3-oxoLCA, isoLCA suppressed TH17 cell differentiation by inhibiting retinoic acid receptor-related orphan nuclear receptor-γt, a key TH17-cell-promoting transcription factor. The levels of both 3-oxoLCA and isoLCA and the 3α-hydroxysteroid dehydrogenase genes that are required for their biosynthesis were significantly reduced in patients with inflammatory bowel disease. Moreover, the levels of these bile acids were inversely correlated with the expression of TH17-cell-associated genes. Overall, our data suggest that bacterially produced bile acids inhibit TH17 cell function, an activity that may be relevant to the pathophysiology of inflammatory disorders such as inflammatory bowel disease.

Strain-level fitness in the gut microbiome is an emergent property of glycans and a single metabolite.

The human gut microbiota resides within a diverse chemical environment challenging our ability to understand the forces shaping this ecosystem. Here, we reveal that fitness of the Bacteroidales, the dominant order of bacteria in the human gut, is an emergent property of glycans and one specific metabolite, butyrate. Distinct sugars serve as strain-variable fitness switches activating context-dependent inhibitory functions of butyrate. Differential fitness effects of butyrate within the Bacteroides are mediated by species-level variation in Acyl-CoA thioesterase activity and nucleotide polymorphisms regulating an Acyl-CoA transferase. Using in vivo multi-omic profiles, we demonstrate Bacteroides fitness in the human gut is associated together, but not independently, with Acyl-CoA transferase expression and butyrate. Our data reveal that each strain of the Bacteroides exists within a unique fitness landscape based on the interaction of chemical components unpredictable by the effect of each part alone mediated by flexibility in the core genome.

Multivariable association discovery in population-scale meta-omics studies.

It is challenging to associate features such as human health outcomes, diet, environmental conditions, or other metadata to microbial community measurements, due in part to their quantitative properties. Microbiome multi-omics are typically noisy, sparse (zero-inflated), high-dimensional, extremely non-normal, and often in the form of count or compositional measurements. Here we introduce an optimized combination of novel and established methodology to assess multivariable association of microbial community features with complex metadata in population-scale observational studies. Our approach, MaAsLin 2 (Microbiome Multivariable Associations with Linear Models), uses generalized linear and mixed models to accommodate a wide variety of modern epidemiological studies, including cross-sectional and longitudinal designs, as well as a variety of data types (e.g., counts and relative abundances) with or without covariates and repeated measurements. To construct this method, we conducted a large-scale evaluation of a broad range of scenarios under which straightforward identification of meta-omics associations can be challenging. These simulation studies reveal that MaAsLin 2's linear model preserves statistical power in the presence of repeated measures and multiple covariates, while accounting for the nuances of meta-omics features and controlling false discovery. We also applied MaAsLin 2 to a microbial multi-omics dataset from the Integrative Human Microbiome (HMP2) project which, in addition to reproducing established results, revealed a unique, integrated landscape of inflammatory bowel diseases (IBD) across multiple time points and omics profiles.

A statistical model for describing and simulating microbial community profiles.

Many methods have been developed for statistical analysis of microbial community profiles, but due to the complex nature of typical microbiome measurements (e.g. sparsity, zero-inflation, non-independence, and compositionality) and of the associated underlying biology, it is difficult to compare or evaluate such methods within a single systematic framework. To address this challenge, we developed SparseDOSSA (Sparse Data Observations for the Simulation of Synthetic Abundances): a statistical model of microbial ecological population structure, which can be used to parameterize real-world microbial community profiles and to simulate new, realistic profiles of known structure for methods evaluation. Specifically, SparseDOSSA's model captures marginal microbial feature abundances as a zero-inflated log-normal distribution, with additional model components for absolute cell counts and the sequence read generation process, microbe-microbe, and microbe-environment interactions. Together, these allow fully known covariance structure between synthetic features (i.e. "taxa") or between features and "phenotypes" to be simulated for method benchmarking. Here, we demonstrate SparseDOSSA's performance for 1) accurately modeling human-associated microbial population profiles; 2) generating synthetic communities with controlled population and ecological structures; 3) spiking-in true positive synthetic associations to benchmark analysis methods; and 4) recapitulating an end-to-end mouse microbiome feeding experiment. Together, these represent the most common analysis types in assessment of real microbial community environmental and epidemiological statistics, thus demonstrating SparseDOSSA's utility as a general-purpose aid for modeling communities and evaluating quantitative methods. An open-source implementation is available at http://huttenhower.sph.harvard.edu/sparsedossa2.