Machine Learning Meta-analysis of Large Metagenomic Datasets: Tools and Biological Insights

Abstract

Shotgun metagenomic analysis of the human associated microbiome provides a rich set of microbial features for prediction and biomarker discovery in the context of human diseases and health conditions. However, the use of such high-resolution microbial features presents new challenges, and validated computational tools for learning tasks are lacking. Moreover, classification rules have scarcely been validated in independent studies, posing questions about the generality and generalization of disease-predictive models across cohorts. In this paper, we comprehensively assess approaches to metagenomics-based prediction tasks and for quantitative assessment of the strength of potential microbiome-phenotype associations. We develop a computational framework for prediction tasks using quantitative microbiome profiles, including species-level relative abundances and presence of strain-specific markers. A comprehensive meta-analysis, with particular emphasis on generalization across cohorts, was performed in a collection of 2424 publicly available metagenomic samples from eight large-scale studies. Cross-validation revealed good disease-prediction capabilities, which were in general improved by feature selection and use of strain-specific markers instead of species-level taxonomic abundance. In cross-study analysis, models transferred between studies were in some cases less accurate than models tested by within-study cross-validation. Interestingly, the addition of healthy (control) samples from other studies to training sets improved disease prediction capabilities. Some microbial species (most notably Streptococcus anginosus) seem to characterize general dysbiotic states of the microbiome rather than connections with a specific disease. Our results in modelling features of the “healthy” microbiome can be considered a first step toward defining general microbial dysbiosis. The software framework, microbiome profiles, and metadata for thousands of samples are publicly available at http://segatalab.cibio.unitn.it/tools/metaml.

Metagenome-Wide Association Study and Machine Learning Prediction of Bulk Soil Microbiome and Crop Productivity

“…The result demonstrated that microbiome composition indeed could be useful for crop productivity prediction. While the prediction model with a small training set resulted in lower accuracy compare to machine learning prediction in human diseases (generally included 100–300 samples; Pasolli et al, 2016), we expect the accuracy would be improved with a larger sample size. Nonetheless, because soil microbiome could be far more complicated and diverse than human microbiome, limited sequencing depth to detect rare taxa and the reproducibility under the challenge of highly varied environmental factors will be the technical bottlenecks.…”

Microbial strain-level population structure and genetic diversity from metagenomes

“…We next applied StrainPhlAn to a set of 1590 gut metagenomes from adult subjects retrieved from nine public data sets ( Table 1) that we preprocessed using uniform quality control criteria (Methods) as in Pasolli et al (2016). The resulting population spanned all continents except Australia and Antarctica, with curated common metadata including country of origin, health or disease state, age, and BMI (other metadata was either not provided or not common among data sets).…”

DeepMicro: deep representation learning for disease prediction based on microbiome data

“…Comparatively, strain level profiles, often containing hundreds of thousands of gene markers' information, should be more informative for accurately classifying the samples into patient and healthy control groups across different types of diseases than abundance profiles that usually contain only a few hundred bacteria's abundance information [11]. Lastly, to evaluate and compare the performance of machine learning models, it is necessary to introduce a rigorous validation framework to estimate their performance over unseen data.…”

“…Lastly, to evaluate and compare the performance of machine learning models, it is necessary to introduce a rigorous validation framework to estimate their performance over unseen data. Pasolli et al utilized a 10-fold cross-validation scheme that optimizes hyper-parameters of their model using a test fold as well as 9 training folds and selects the best result as its performance [11]. This approach may overestimate model performance as it allows to tune the model against test data [12,13].…”