Accurate identification of bacteriophages from metagenomic data using Transformer

Abstract

Motivation Bacteriophages are viruses infecting bacteria. Being key players in microbial communities, they can regulate the composition/function of microbiome by infecting their bacterial hosts and mediating gene transfer. Recently, metagenomic sequencing, which can sequence all genetic materials from various microbiome, has become a popular means for new phage discovery. However, accurate and comprehensive detection of phages from the metagenomic data remains difficult. High diversity/abundance, and limited reference genomes pose major challenges for recruiting phage fragments from metagenomic data. Existing alignment-based or learning-based models have either low recall or precision on metagenomic data. Results In this work, we adopt the state-of-the-art language model, Transformer, to conduct contextual embedding for phage contigs. By constructing a protein-cluster vocabulary, we can feed both the protein composition and the proteins’ positions from each contig into the Transformer. The Transformer can learn the protein organization and associations using the self-attention mechanism and predicts the label for test contigs. We rigorously tested our developed tool named PhaMer on multiple datasets with increasing difficulty, including quality RefSeq genomes, short contigs, simulated metagenomic data, mock metagenomic data and the public IMG/VR dataset. All the experimental results show that PhaMer outperforms the state-of-the-art tools. In the real metagenomic data experiment, PhaMer improves the F1-score of phage detection by 27%.

Evaluation of computational phage detection tools for metagenomic datasets

“…Despite these limitations, we hope the developed benchmark may be informative to users and would be further developed to include new computational challenges. It should be noted that the results presented here are limited to those tools that could be installed and run by July 2021, and since then many more tools have been published [we are aware of 3CAC ( Pu and Shamir, 2022 ), DeepMicrobeFinder ( Hou et al, 2021 ), INHERIT ( Bai et al, 2022 ), PHAMB ( Johansen et al, 2022 ), PhaMer ( Shang et al, 2022 ), VirMine 2.0 ( Johnson and Putonti, 2022 ), and virSearcher ( Liu Q. et al, 2022 )]. Additionally, modular pipelines such as the IMG/VR viral discovery pipeline ( Paez-Espino et al, 2017 ) and computational pipelines combining several tools presented here, were not evaluated in this work but could be assessed using the same benchmark datasets developed here.…”

Phage family classification under Caudoviricetes: A review of current tools using the latest ICTV classification framework

“…Thus, a pre-processing step is needed for detecting those contigs from metagenomic data. A number of tools, such as VirFinder (Ren et al, 2020), Seeker (Auslander et al, 2020), and PhaMer (Shang et al, 2022) can be applied in the pre-processing step.…”

“…• Simulated metagenomic dataset We used a simulated metagenomic dataset generated by six common bacteria living in human gut (Shang et al, 2022). We first utilized metaSPAdes (Nurk et al, 2017) to assemble the reads into contigs.…”

“…We first utilized metaSPAdes (Nurk et al, 2017) to assemble the reads into contigs. Then PhaMer (Shang et al, 2022) was applied to identify bacteriophages from metagenomic data, and the labels of the contigs were determined using BLAST (Camacho et al, 2009). Eventually, 37 contigs were used in the experiments.…”

“…In this experiment, we used the simulated metagenomic dataset provided in PhaMer (Shang et al, 2022). The dataset is a small-scale metagenomic dataset simulated by CAMISIM (Fritz et al, 2019) using the commonly seen bacteria living in the human gut and the phages that infect these bacteria.…”