MiMeNet: Exploring Microbiome-Metabolome Relationships using Neural Networks

Abstract

The advance in microbiome and metabolome studies has generated rich omics data revealing the involvement of the microbial community in host disease pathogenesis through interactions with their host at a metabolic level. However, the computational tools to uncover these relationships are just emerging. Here, we present MiMeNet, a neural network framework for modeling microbe-metabolite relationships. Using ten iterations of 10-fold cross-validation on three paired microbiome-metabolome datasets, we show that MiMeNet more accurately predicts metabolite abundances (mean Spearman correlation coefficients increase from 0.108 to 0.309, 0.276 to 0.457, and -0.272 to 0.264) and identifies more well-predicted metabolites (increase in the number of well-predicted metabolites from 198 to 366, 104 to 143, and 4 to 29) compared to state-of-art linear models for individual metabolite predictions. Additionally, we demonstrate that MiMeNet can group microbes and metabolites with similar interaction patterns and functions to illuminate the underlying structure of the microbe-metabolite interaction network, which could potentially shed light on uncharacterized metabolites through "Guilt by Association". Our results demonstrated that MiMeNet is a powerful tool to provide insights into the causes of metabolic dysregulation in disease, facilitating future hypothesis generation at the interface of the microbiome and metabolomics.

Do amino and fatty acid profiles of pollen provisions correlate with bacterial microbiomes in the mason beeOsmia bicornis?

“…We were further interested in links between the relative abundance of individual bacterial community members and the centred log-ratio transformed abundance of individual amino acids and fatty acids in pollen provisions. For this, we used multi-layer perceptron neural networks using MiMeNet [91]. This method has been specifically designed to address microbe–metabolite interactions and can identify significantly correlated nutrient–bacteria pairs by testing against null-model background distributions.…”

“…This method has been specifically designed to address microbe–metabolite interactions and can identify significantly correlated nutrient–bacteria pairs by testing against null-model background distributions. It has been evaluated to be robust with respect to dataset size [91], which was important for our comparisons (provision bacterial communities versus provision nutrients: 21 AA and 19 FA samples from the same nest cells). For each analysis, 10 models were trained and validated in a 8 : 2 ratio with 10 cross-validation runs, each with 10 folds, and compared with 10 background shuffled cross-validations to identify Spearman rank correlation coefficients between bacterial genera and nutrients.…”

Predicting metabolomic profiles from microbial composition through neural ordinary differential equations

“…Figs. 2a1-a3 show the performance comparison between mNODE and previous ML-based methods (MelonnPan [26], sparse NED [27], MiMeNet [28], and ResNet [35]) through 3 metrics: (1) mean SCC, (Figs. 2a1, b1, c1), (2) the mean of the top-5 SCCs, (Figs.…”

“…(3) Machine learning (ML)-based methods, which are trained from paired microbiome and metabolome datasets, and then used to predict the metabolic profile of a never-seen microbiome sample based on its microbial composition, without using any reference database or domain knowledge regarding relationships between genes and metabolites. Various ML techniques such as elastic net [26], sparsified NED (Neural Encoder-Decoder) [27], multilayer perceptron [28], and word2vec [29] have been employed to predict metabolic profiles from microbial compositions.…”

“…(3) Machine learning (ML)-based methods, which are trained from paired microbiome and metabolome datasets, and then used to predict the metabolic profile of a never-seen microbiome sample based on its microbial composition, without using any reference database or domain knowledge regarding relationships between genes and metabolites. Various ML techniques such as elastic net [26], sparsified NED (Neural Encoder-Decoder) [27], multilayer perceptron [28], and word2vec [29] have been employed to predict metabolic profiles from microbial compositions. Despite the fact that these ML-based methods circumvent limitations of reference-based or ecology-guided methods discussed above and have been shown to generate promising results in various contexts, none of these ML-based methods utilize state-of-the-art deep neural network models such as the Neural Ordinary Differential Equation (NODE) [30], so their performance has not been fully maximized.…”