Improved metagenome binning and assembly using deep variational autoencoders

Abstract

Identification and reconstruction of microbial species from metagenomics wide genome sequencing data is an important and challenging task. Current existing approaches rely on gene or contig co-abundance information across multiple samples and k -mer composition information in the sequences. Here we use recent advances in deep learning to develop an algorithm that uses variational autoencoders to encode co-abundance and compositional information prior to clustering. We show that the deep network is able to integrate these two heterogeneous datasets without any prior knowledge and that our method outperforms existing state-of-the-art by reconstructing 1.8 -8 times more highly precise and complete genome bins from three different benchmark datasets. Additionally, we apply our method to a gene catalogue of almost 10 million genes and 1,270 samples from the human gut microbiome. Here we are able to cluster 1.3 -1.8 million extra genes and reconstruct 117 -246 more highly precise and complete bins of which 70 bins were completely new compared to previous methods. Our method Variational Autoencoders for Metagenomic Binning (VAMB) is freely available at: https://github.com/jakobnissen/vamb

MetaBinner: a high-performance and stand-alone ensemble binning method to recover individual genomes from complex microbial communities

“…In the following, we compared the performance of Metabinner with other individual binners (CONCOCT [12], MetaBAT [34,13], MaxBin [14,15], VAMB [16]) and ensemble binners (BMC3C [22], MetaWRAP [21] and DAS Tool [19]). Then, we conducted experiments to show the necessity and effectiveness of multiple features and initializations.…”

“…MaxBin [14,15] multiplies the probability P dist and the probability P cov that a sequence belongs to a bin based on nucleotide frequency distance and coverage, respectively. A deep learning-based binner, VAMB [16], has recently been developed, which utilizes variational autoencoders (VAE) [17] to convert nucleotide information and coverage information for binning. VAMB then clusters the transformed data using an adaptive iterative medoid method.…”

Precise genotyping of circular mobile elements uncovers human associated plasmids with surprisingly recent common ancestors

“…Combining long-reads with the approach presented here will allow genotyping of complex ecMGEs, such as MGEs that contain insertion sequences (typically <2.5kb) and short transposons. Longer repeat elements and complex rearrangements that require bridging over more than ~10kb can be addressed with additional experimental work, such as Hi- 28,[36][37][38] and by sampling the same community multiple times 39,40 .…”

Analysis of viromes and microbiomes from pig fecal samples reveals that phages and prophages are not vectors of antibiotic resistance genes

“…To estimate the main phyla present in the microbiome samples, a taxonomic affiliation was determined using Kaiju 25 for the top 10 585 P contigs with a global abundance above 10 RPKM. To sum up contig abundances by phyla, those which might belong to the same microbial species were binned into clusters based on kmer composition and co-occurrence with VAMB 26 . A single contig per VAMB cluster was conserved in the abundance matrix, which ended up containing 5862 contigs (~species).…”