G2S: A New Deep Learning Tool for Predicting Stool Microbiome Structure From Oral Microbiome Data

Abstract

Deep learning methodologies have revolutionized prediction in many fields and show the potential to do the same in microbial metagenomics. However, deep learning is still unexplored in the field of microbiology, with only a few software designed to work with microbiome data. Within the meta-community theory, we foresee new perspectives for the development and application of deep learning algorithms in the field of the human microbiome. In this context, we developed G2S, a bioinformatic tool for taxonomic prediction of the human fecal microbiome directly from the oral microbiome data of the same individual. The tool uses a deep convolutional neural network trained on paired oral and fecal samples from populations across the globe, which allows inferring the stool microbiome at the family level more accurately than other available approaches. The tool can be used in retrospective studies, where fecal sampling was not performed, and especially in the field of paleomicrobiology, as a unique opportunity to recover data related to ancient gut microbiome configurations. G2S was validated on already characterized oral and fecal sample pairs, and then applied to ancient microbiome data from dental calculi, to derive putative intestinal components in medieval subjects.

Keywords: deep learning; gut microbiome; microbiome; oral microbiome; paleomicrobiology.

FIGURE 1

G2S workflow. The input file is a genus-level relative abundance table (.tsv format), obtained from the characterization of human oral microbiome samples. The stool microbiome is inferred using a deep convolutional neural network (ConvNet) adjusted by a confusion matrix and rescaled to 100%. The results are tabulated as relative abundance.

FIGURE 2

Comparison between G2S predictions and real data from the test dataset. The family level bar plots of the 79 stool samples of the test dataset are visualized next to their inferred configurations obtained by G2S. Spearman correlation coefficients (r) are provided below each pair of bar plots. Samples are derived from the following studies: The Human Microbiome Project Consortium (2012), Zaura et al. (2015); Brito et al. (2016), Russo et al. (2018).

FIGURE 3

G2S predictions are more accurate when the configurations to be inferred fall within the plane of variation of the training dataset. Box plots of the mean absolute error scaled to one standard deviation (maes) between the real stool microbiome configuration of the samples in the test dataset and the median configuration of the training dataset. Samples were divided into four groups based on the quality of the G2S predictions (i.e., the Spearman correlation coefficients between the real values and the inferred configurations).

FIGURE 4

G2S predicts the stool microbiome configuration with better performance than other methods. The mean absolute errors scaled to one standard deviation (maes) between the real data of the samples from the test dataset and the configurations inferred by G2S, Random Forest and a stochastic permutational method (100 predictions), are reported in the dot plot.

FIGURE 5

Reconstructing the ancient stool microbiome of adult medieval individuals. (A) Bar plots of stool microbiome configurations inferred from 16S rRNA gene (V5 and V6 regions) sequencing data of ancient microbiomes (i.e., dental calculi from the medieval monastic site of Dalheim, Germany [ca. 950–1,200 CE]) (Warinner et al., 2014). (B) Comparison between the predicted ancient microbiome configurations and the modern stool microbiome of subjects from the dataset used to implement G2S (The Human Microbiome Project Consortium, 2012; Zaura et al., 2015; Brito et al., 2016; Russo et al., 2018). P-values were determined by Wilcoxon test.