Biological language model performance depends heavily on pretraining data quality, diversity, and size. While metagenomic datasets feature enormous biological diversity, their utilization as pretraining data has been limited due to challenges in data accessibility, quality filtering and deduplication. Here, we present the Open MetaGenomic (OMG) corpus, a genomic pretraining dataset totalling 3.1T base pairs and 3.3B protein coding sequences, obtained by combining two largest metagenomic dataset repositories (JGI's IMG and EMBL's MGnify). We first document the composition of the dataset and describe the quality filtering steps taken to remove poor quality data. We make the OMG corpus available as a mixed-modality genomic sequence dataset that represents multi-gene encoding genomic sequences with translated amino acids for protein coding sequences, and nucleic acids for intergenic sequences. We train the first mixed-modality genomic language model (gLM2) that leverages genomic context information to learn robust functional representations, as well as coevolutionary signals in protein-protein interfaces and genomic regulatory syntax. Furthermore, we show that deduplication in embedding space can be used to balance the corpus, demonstrating improved performance on downstream tasks. The OMG dataset is publicly hosted on the Hugging Face Hub at https://huggingface.co/datasets/tattabio/OMG and gLM2 is available at https://huggingface.co/tattabio/gLM2_650M.

SUMMARY: In recent years, there has been a surge in prokaryotic genome assemblies, coming from both isolated organisms and environmental samples. These assemblies often include novel species that are poorly represented in reference databases creating a need for a tool that can annotate both well-described and novel taxa, and can run at scale. Here, we present mettannotator-a comprehensive, scalable Nextflow pipeline for prokaryotic genome annotation that identifies coding and noncoding regions, predicts protein functions, including antimicrobial resistance, and delineates gene clusters. The pipeline summarizes these results in a GFF (General Feature Format) file that can be easily utilized in downstream analysis or visualized using common genome browsers. Here, we show how it works on 200 genomes from 29 prokaryotic phyla, including isolate genomes and known and novel metagenome-assembled genomes, and present metrics on its performance in comparison to other tools. AVAILABILITY AND IMPLEMENTATION: The pipeline is written in Nextflow and Python and published under an open source Apache 2.0 licence. Instructions and source code can be accessed at https://github.com/EBI-Metagenomics/mettannotator. The pipeline is also available on WorkflowHub: https://workflowhub.eu/workflows/1069.

The study of viral communities has revealed the enormous diversity and impact these biological entities have on various ecosystems. These observations have sparked widespread interest in developing computational strategies that support the comprehensive characterisation of viral communities based on sequencing data. Here we introduce VIRify, a new computational pipeline designed to provide a user-friendly and accurate functional and taxonomic characterisation of viral communities. VIRify identifies viral contigs and prophages from metagenomic assemblies and annotates them using a collection of viral profile hidden Markov models (HMMs). These include our manually-curated profile HMMs, which serve as specific taxonomic markers for a wide range of prokaryotic and eukaryotic viral taxa and are thus used to reliably classify viral contigs. We tested VIRify on assemblies from two microbial mock communities, a large metagenomics study, and a collection of publicly available viral genomic sequences from the human gut. The results showed that VIRify could identify sequences from both prokaryotic and eukaryotic viruses, and provided taxonomic classifications from the genus to the family rank with an average accuracy of 86.6%. In addition, VIRify allowed the detection and taxonomic classification of a range of prokaryotic and eukaryotic viruses present in 243 marine metagenomic assemblies. Finally, the use of VIRify led to a large expansion in the number of taxonomically classified human gut viral sequences and the improvement of outdated and shallow taxonomic classifications. Overall, we demonstrate that VIRify is a novel and powerful resource that offers an enhanced capability to detect a broad range of viral contigs and taxonomically classify them.