Hierarchical Microbial Functions Prediction by Graph Aggregated Embedding.

Yujie Hou,Wenxing Hong,Xiong Zhang,Qinyan Zhou,Ying Wang

doi:10.3389/fgene.2020.608512

Abstract

Matching 16S rRNA gene sequencing data to a metabolic reference database is a meaningful way to predict the metabolic function of bacteria and archaea, bringing greater insight to the working of the microbial community. However, some operational taxonomy units (OTUs) cannot be functionally profiled, especially for microbial communities from non-human samples cultured in defective media. Therefore, we herein report the development of Hierarchical micrObial functions Prediction by graph aggregated Embedding (HOPE), which utilizes co-occurring patterns and nucleotide sequences to predict microbial functions. HOPE integrates topological structures of microbial co-occurrence networks with k-mer compositions of OTU sequences and embeds them into a lower-dimensional continuous latent space, while maximally preserving topological relationships among OTUs. The high imbalance among KEGG Orthology (KO) functions of microbes is recognized in our framework that usually yields poor performance. A hierarchical multitask learning module is used in HOPE to alleviate the challenge brought by the long-tailed distribution among classes. To test the performance of HOPE, we compare it with HOPE-one, HOPE-seq, and GraphSAGE, respectively, in three microbial metagenomic 16s rRNA sequencing datasets, including abalone gut, human gut, and gut of Penaeus monodon. Experiments demonstrate that HOPE outperforms baselines on almost all indexes in all experiments. Furthermore, HOPE reveals significant generalization ability. HOPE's basic idea is suitable for other related scenarios, such as the prediction of gene function based on gene co-expression networks. The source code of HOPE is freely available at https://github.com/adrift00/HOPE.

Highlights

The analysis of microbial communities is founded on the characterization of functional diversity, which is increasingly recognized as the bridge linking biodiversity patterns and ecosystem functioning, as a way of explaining the interactions between microbes and their responses to changes in the environment (Bardgett and Der Putten, 2014; Escalas et al, 2019)
The second strategy trains the model on one type of microbial community and tests that model on a different, but related, microbial community, aiming to test the generalization ability of HOPE
Discussion for Class Imbalance Problem We find that KEGG Orthology (KO) functions with a large number of annotation samples generally outperform KO functions with a few annotations

Summary

Introduction

The analysis of microbial communities is founded on the characterization of functional diversity, which is increasingly recognized as the bridge linking biodiversity patterns and ecosystem functioning, as a way of explaining the interactions between microbes and their responses to changes in the environment (Bardgett and Der Putten, 2014; Escalas et al, 2019). Because of the prevalence of high-throughput sequencing technologies, large-scale 16S rRNA marker gene sequencing of microbes is becoming available Related approaches, such as PICRUSt (Langille et al, 2013) and Tax4Fun (Ashauer et al, 2015), are proposed to infer functional profiles from genomes and phylogeny. PICRUSt and Tax4Fun identify microbial functions by estimating 16s rRNA marker gene families based on the similarity between 16s rRNA sequencing data and known marker gene databases. They rely on the reference databases on Greengenes (Desantis et al, 2006) and SILVA (Quast et al, 2012). Owing to the incompleteness of the 16S rRNA marker gene database, large amounts of OTUs cannot be functionally profiled, especially for microbial communities from non-human samples from defective culture media (Pachiadaki et al, 2019; Wang X. et al, 2020)

Methods

Results

Discussion

Conclusion