A systematic machine learning and data type comparison yields metagenomic predictors of infant age, sex, breastfeeding, antibiotic usage, country of origin, and delivery type

Multi-omic integration of microbiome data for identifying disease-associated modules

The human gut microbiome is a complex ecosystem with profound implications for health and disease. This recognition has led to a surge in multi-omic microbiome studies, employing various molecular assays to elucidate the microbiome's role in diseases across multiple functional layers. However, despite the clear value of these multi-omic datasets, rigorous integrative analysis of such data poses significant challenges, hindering a comprehensive understanding of microbiome-disease interactions. Perhaps most notably, multiple approaches, including univariate and multivariate analyses, as well as machine learning, have been applied to such data to identify disease-associated markers, namely, specific features (e.g., species, pathways, metabolites) that are significantly altered in disease state. These methods, however, often yield extensive lists of features associated with the disease without effectively capturing the multi-layered structure of multi-omic data or offering clear, interpretable hypotheses about underlying microbiome-disease mechanisms. Here, we address this challenge by introducing MintTea - an intermediate integration-based method for analyzing multi-omic microbiome data. MintTea combines a canonical correlation analysis (CCA) extension, consensus analysis, and an evaluation protocol to robustly identify disease-associated multi-omic modules. Each such module consists of a set of features from the various omics that both shift in concord, and collectively associate with the disease. Applying MintTea to diverse case-control cohorts with multi-omic data, we show that this framework is able to capture modules with high predictive power for disease, significant cross-omic correlations, and alignment with known microbiome-disease associations. For example, analyzing samples from a metabolic syndrome (MS) study, we found a MS-associated module comprising of a highly correlated cluster of serum glutamate- and TCA cycle-related metabolites, as well as bacterial species previously implicated in insulin resistance. In another cohort, we identified a module associated with late-stage colorectal cancer, featuring Peptostreptococcus and Gemella species and several fecal amino acids, in agreement with these species' reported role in the metabolism of these amino acids and their coordinated increase in abundance during disease development. Finally, comparing modules identified in different datasets, we detected multiple significant overlaps, suggesting common interactions between microbiome features. Combined, this work serves as a proof of concept for the potential benefits of advanced integration methods in generating integrated multi-omic hypotheses underlying microbiome-disease interactions and a promising avenue for researchers seeking systems-level insights into coherent mechanisms governing microbiome-related diseases.

Comprehensive assessment of machine learning methods for diagnosing gastrointestinal diseases through whole metagenome sequencing data

The gut microbiome, linked significantly to host diseases, offers potential for disease diagnosis through machine learning (ML) pipelines. These pipelines, crucial in modeling diseases using high-dimensional microbiome data, involve selecting profile modalities, data preprocessing techniques, and classification algorithms, each impacting the model accuracy and generalizability. Despite whole metagenome shotgun sequencing (WMS) gaining popularity for human gut microbiome profiling, a consensus on the optimal methods for ML pipelines in disease diagnosis using WMS data remains elusive. Addressing this gap, we comprehensively evaluated ML methods for diagnosing Crohn's disease and colorectal cancer, using 2,553 fecal WMS samples from 21 case-control studies. Our study uncovered crucial insights: gut-specific, species-level taxonomic features proved to be the most effective for profiling; batch correction was not consistently beneficial for model performance; compositional data transformations markedly improved the models; and while nonlinear ensemble classification algorithms typically offered superior performance, linear models with proper regularization were found to be more effective for diseases that are linearly separable based on microbiome data. An optimal ML pipeline, integrating the most effective methods, was validated for generalizability using holdout data. This research offers practical guidelines for constructing reliable disease diagnostic ML models with fecal WMS data.

Incorporating metabolic activity, taxonomy and community structure to improve microbiome-based predictive models for host phenotype prediction

We developed MicroKPNN, a prior-knowledge guided interpretable neural network for microbiome-based human host phenotype prediction. The prior knowledge used in MicroKPNN includes the metabolic activities of different bacterial species, phylogenetic relationships, and bacterial community structure, all in a shallow neural network. Application of MicroKPNN to seven gut microbiome datasets (involving five different human diseases including inflammatory bowel disease, type 2 diabetes, liver cirrhosis, colorectal cancer, and obesity) shows that incorporation of the prior knowledge helped improve the microbiome-based host phenotype prediction. MicroKPNN outperformed fully connected neural network-based approaches in all seven cases, with the most improvement of accuracy in the prediction of type 2 diabetes. MicroKPNN outperformed a recently developed deep-learning based approach DeepMicro, which selects the best combination of autoencoder and machine learning approach to make predictions, in all of the seven cases. Importantly, we showed that MicroKPNN provides a way for interpretation of the predictive models. Using importance scores estimated for the hidden nodes, MicroKPNN could provide explanations for prior research findings by highlighting the roles of specific microbiome components in phenotype predictions. In addition, it may suggest potential future research directions for studying the impacts of microbiome on host health and diseases. MicroKPNN is publicly available at https://github.com/mgtools/MicroKPNN.

Machine-learning analysis of cross-study samples according to the gut microbiome in 12 infant cohorts

IMPORTANCE

There are challenges in merging microbiome data from diverse research groups due to the intricate and multifaceted nature of such data. To address this, we utilized a combination of machine-learning (ML) models to analyze 16S sequencing data from a substantial set of gut microbiome samples, sourced from 12 distinct infant cohorts that were gathered prospectively. Our initial focus was on the mode of delivery due to its prior association with changes in infant gut microbiomes. Through ML analysis, we demonstrated the effective merging and comparison of various gut microbiome data sets, facilitating the identification of robust microbiome biomarkers applicable across varied study populations.

Unveiling the Connection between Microbiota and Depressive Disorder through Machine Learning

In the last few years, investigation of the gut-brain axis and the connection between the gut microbiota and the human nervous system and mental health has become one of the most popular topics. Correlations between the taxonomic and functional changes in gut microbiota and major depressive disorder have been shown in several studies. Machine learning provides a promising approach to analyze large-scale metagenomic data and identify biomarkers associated with depression. In this work, machine learning algorithms, such as random forest, elastic net, and You Only Look Once (YOLO), were utilized to detect significant features in microbiome samples and classify individuals based on their disorder status. The analysis was conducted on metagenomic data obtained during the study of gut microbiota of healthy people and patients with major depressive disorder. The YOLO method showed the greatest effectiveness in the analysis of the metagenomic samples and confirmed the experimental results on the critical importance of a reduction in the amount of Faecalibacterium prausnitzii for the manifestation of depression. These findings could contribute to a better understanding of the role of the gut microbiota in major depressive disorder and potentially lead the way for novel diagnostic and therapeutic strategies.

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.

Quantifying Shared and Unique Gene Content across 17 Microbial Ecosystems

Measuring microbial diversity is traditionally based on microbe taxonomy. Here, in contrast, we aimed to quantify heterogeneity in microbial gene content across 14,183 metagenomic samples spanning 17 ecologies, including 6 human associated, 7 nonhuman host associated, and 4 in other nonhuman host environments. In total, we identified 117,629,181 nonredundant genes. The vast majority of genes (66%) occurred in only one sample (i.e., "singletons"). In contrast, we found 1,864 sequences present in every metagenome, but not necessarily every bacterial genome. Additionally, we report data sets of other ecology-associated genes (e.g., abundant in only gut ecosystems) and simultaneously demonstrated that prior microbiome gene catalogs are both incomplete and inaccurately cluster microbial genetic life (e.g., at gene sequence identities that are too restrictive). We provide our results and the sets of environmentally differentiating genes described above at http://www.microbial-genes.bio. IMPORTANCE The amount of shared genetic elements has not been quantified between the human microbiome and other host- and non-host-associated microbiomes. Here, we made a gene catalog of 17 different microbial ecosystems and compared them. We show that most species shared between environment and human gut microbiomes are pathogens and that prior gene catalogs described as "nearly complete" are far from it. Additionally, over two-thirds of all genes only appear in a single sample, and only 1,864 genes (0.001%) are found in all types of metagenomes. These results highlight the large diversity between metagenomes and reveal a new, rare class of genes, those found in every type of metagenome, but not every microbial genome.

Supervised Machine Learning Enables Geospatial Microbial Provenance

The recent increase in publicly available metagenomic datasets with geospatial metadata has made it possible to determine location-specific, microbial fingerprints from around the world. Such fingerprints can be useful for comparing microbial niches for environmental research, as well as for applications within forensic science and public health. To determine the regional specificity for environmental metagenomes, we examined 4305 shotgun-sequenced samples from the MetaSUB Consortium dataset-the most extensive public collection of urban microbiomes, spanning 60 different cities, 30 countries, and 6 continents. We were able to identify city-specific microbial fingerprints using supervised machine learning (SML) on the taxonomic classifications, and we also compared the performance of ten SML classifiers. We then further evaluated the five algorithms with the highest accuracy, with the city and continental accuracy ranging from 85-89% to 90-94%, respectively. Thereafter, we used these results to develop Cassandra, a random-forest-based classifier that identifies bioindicator species to aid in fingerprinting and can infer higher-order microbial interactions at each site. We further tested the Cassandra algorithm on the Tara Oceans dataset, the largest collection of marine-based microbial genomes, where it classified the oceanic sample locations with 83% accuracy. These results and code show the utility of SML methods and Cassandra to identify bioindicator species across both oceanic and urban environments, which can help guide ongoing efforts in biotracing, environmental monitoring, and microbial forensics (MF).

From the nose to the lungs: the intricate journey of airborne pathogens amidst commensal bacteria

The recent COVID-19 pandemic has dramatically brought the pitfalls of airborne pathogens to the attention of the scientific community. Not only viruses but also bacteria and fungi may exploit air transmission to colonize and infect potential hosts and be the cause of significant morbidity and mortality in susceptible populations. The efforts to decipher the mechanisms of pathogenicity of airborne microbes have brought to light the delicate equilibrium that governs the homeostasis of mucosal membranes. The microorganisms already thriving in the permissive environment of the respiratory tract represent a critical component of this equilibrium and a potent barrier to infection by means of direct competition with airborne pathogens or indirectly via modulation of the immune response. Moving down the respiratory tract, physicochemical and biological constraints promote site-specific expansion of microbes that engage in cross talk with the local immune system to maintain homeostasis and promote protection. In this review, we critically assess the site-specific microbial communities that an airborne pathogen encounters in its hypothetical travel along the respiratory tract and discuss the changes in the composition and function of the microbiome in airborne diseases by taking fungal and SARS-CoV-2 infections as examples. Finally, we discuss how technological and bioinformatics advancements may turn microbiome analysis into a valuable tool in the hands of clinicians to predict the risk of disease onset, the clinical course, and the response to treatment of individual patients in the direction of personalized medicine implementation.

Using deep learning to predict abdominal age from liver and pancreas magnetic resonance images

With age, the prevalence of diseases such as fatty liver disease, cirrhosis, and type two diabetes increases. Approaches to both predict abdominal age and identify risk factors for accelerated abdominal age may ultimately lead to advances that will delay the onset of these diseases. We build an abdominal age predictor by training convolutional neural networks to predict abdominal age (or "AbdAge") from 45,552 liver magnetic resonance images [MRIs] and 36,784 pancreas MRIs (R-Squared = 73.3 ± 0.6; mean absolute error = 2.94 ± 0.03 years). Attention maps show that the prediction is driven by both liver and pancreas anatomical features, and surrounding organs and tissue. Abdominal aging is a complex trait, partially heritable (h_g² = 26.3 ± 1.9%), and associated with 16 genetic loci (e.g. in PLEKHA1 and EFEMP1), biomarkers (e.g body impedance), clinical phenotypes (e.g, chest pain), diseases (e.g. hypertension), environmental (e.g smoking), and socioeconomic (e.g education, income) factors.

Systematically assessing microbiome–disease associations identifies drivers of inconsistency in metagenomic research

Evaluating the relationship between the human gut microbiome and disease requires computing reliable statistical associations. Here, using millions of different association modeling strategies, we evaluated the consistency—or robustness—of microbiome-based disease indicators for 6 prevalent and well-studied phenotypes (across 15 public cohorts and 2,343 individuals). We were able to discriminate between analytically robust versus nonrobust results. In many cases, different models yielded contradictory associations for the same taxon–disease pairing, some showing positive correlations and others negative. When querying a subset of 581 microbe–disease associations that have been previously reported in the literature, 1 out of 3 taxa demonstrated substantial inconsistency in association sign. Notably, >90% of published findings for type 1 diabetes (T1D) and type 2 diabetes (T2D) were particularly nonrobust in this regard. We additionally quantified how potential confounders—sequencing depth, glucose levels, cholesterol, and body mass index, for example—influenced associations, analyzing how these variables affect the ostensible correlation between Faecalibacterium prausnitzii abundance and a healthy gut. Overall, we propose our approach as a method to maximize confidence when prioritizing findings that emerge from microbiome association studies.

The human microbiome has been associated with many aspects of our health, but how many of these associations are truly reproducible? This study attempts to address this question by systematically testing the robustness of 581 microbial features that have been reported as being disease-associated.

Human gut microbiome aging clocks based on taxonomic and functional signatures through multi-view learning

The human gut microbiome is a complex ecosystem that is closely related to the aging process. However, there is currently no reliable method to make full use of the metagenomics data of the gut microbiome to determine the age of the host. In this study, we considered the influence of geographical factors on the gut microbiome, and a total of 2604 filtered metagenomics data from the gut microbiome were used to construct an age prediction model. Then, we developed an ensemble model with multiple heterogeneous algorithms and combined species and pathway profiles for multi-view learning. By integrating gut microbiome metagenomics data and adjusting host confounding factors, the model showed high accuracy (R² = 0.599, mean absolute error = 8.33 years). Besides, we further interpreted the model and identify potential biomarkers for the aging process. Among these identified biomarkers, we found that Finegoldia magna, Bifidobacterium dentium, and Clostridium clostridioforme had increased abundance in the elderly. Moreover, the utilization of amino acids by the gut microbiome undergoes substantial changes with increasing age which have been reported as the risk factors for age-associated malnutrition and inflammation. This model will be helpful for the comprehensive utilization of multiple omics data, and will allow greater understanding of the interaction between microorganisms and age to realize the targeted intervention of aging.

Delivery mode and perinatal antibiotics influence the predicted metabolic pathways of the gut microbiome

Delivery mode and perinatal antibiotics influence gut microbiome composition in children. Most microbiome studies have used the sequencing of the bacterial 16S marker gene but have not reported the metabolic function of the gut microbiome, which may mediate biological effects on the host. Here, we used the PICRUSt2 bioinformatics tool to predict the functional profiles of the gut microbiome based on 16S sequencing in two child cohorts. Both Caesarean section and perinatal antibiotics markedly influenced the functional profiles of the gut microbiome at the age of 1 year. In machine learning analysis, bacterial fatty acid, phospholipid, and biotin biosynthesis were the most important pathways that differed according to delivery mode. Proteinogenic amino acid biosynthesis, carbohydrate degradation, pyrimidine deoxyribonucleotide and biotin biosynthesis were the most important pathways differing according to antibiotic exposure. Our study shows that both Caesarean section and perinatal antibiotics markedly influence the predicted metabolic profiles of the gut microbiome at the age of 1 year.

Microbiome meta-analysis and cross-disease comparison enabled by the SIAMCAT machine learning toolbox

The human microbiome is increasingly mined for diagnostic and therapeutic biomarkers using machine learning (ML). However, metagenomics-specific software is scarce, and overoptimistic evaluation and limited cross-study generalization are prevailing issues. To address these, we developed SIAMCAT, a versatile R toolbox for ML-based comparative metagenomics. We demonstrate its capabilities in a meta-analysis of fecal metagenomic studies (10,803 samples). When naively transferred across studies, ML models lost accuracy and disease specificity, which could however be resolved by a novel training set augmentation strategy. This reveals some biomarkers to be disease-specific, with others shared across multiple conditions. SIAMCAT is freely available from siamcat.embl.de .

Applications of Machine Learning in Human Microbiome Studies: A Review on Feature Selection, Biomarker Identification, Disease Prediction and Treatment

The number of microbiome-related studies has notably increased the availability of data on human microbiome composition and function. These studies provide the essential material to deeply explore host-microbiome associations and their relation to the development and progression of various complex diseases. Improved data-analytical tools are needed to exploit all information from these biological datasets, taking into account the peculiarities of microbiome data, i.e., compositional, heterogeneous and sparse nature of these datasets. The possibility of predicting host-phenotypes based on taxonomy-informed feature selection to establish an association between microbiome and predict disease states is beneficial for personalized medicine. In this regard, machine learning (ML) provides new insights into the development of models that can be used to predict outputs, such as classification and prediction in microbiology, infer host phenotypes to predict diseases and use microbial communities to stratify patients by their characterization of state-specific microbial signatures. Here we review the state-of-the-art ML methods and respective software applied in human microbiome studies, performed as part of the COST Action ML4Microbiome activities. This scoping review focuses on the application of ML in microbiome studies related to association and clinical use for diagnostics, prognostics, and therapeutics. Although the data presented here is more related to the bacterial community, many algorithms could be applied in general, regardless of the feature type. This literature and software review covering this broad topic is aligned with the scoping review methodology. The manual identification of data sources has been complemented with: (1) automated publication search through digital libraries of the three major publishers using natural language processing (NLP) Toolkit, and (2) an automated identification of relevant software repositories on GitHub and ranking of the related research papers relying on learning to rank approach.