Libusha Kelly
3.9K posts

Libusha Kelly
@microbegrrl
Microbial ecologist, computational and systems biologist, genomics freak







Improving viral annotation with artificial intelligence 1/ This review highlights the potential of using artificial intelligence, specifically protein language models (pLMs), to tackle one of virology's toughest challenges: the annotation of viral sequences that lack homology with known proteins. 2/ The key innovation is leveraging pLMs to bridge the gap left by traditional sequence-based annotation tools, which struggle with the diversity and rapid evolution of viral proteins, especially those from uncultivated viral genomes. 3/ A central breakthrough is the use of self-supervised learning to train models on massive datasets of amino acid sequences. These models capture structural and functional properties of viral proteins, even when sequence homology is too distant to detect. 4/ The paper emphasizes how protein embeddings generated by pLMs can predict functions for unknown viral proteins, significantly improving annotation for metagenomic data, especially for ocean viromes. 5/ One notable success is the expansion of annotated viral protein families using pretrained pLMs, discovering novel viral capsid proteins and mobile genetic elements previously hidden in environmental samples. 6/ The review outlines key limitations of current pLMs, including their bias toward well-represented protein families and the computational resources required for training. However, it proposes future directions, such as fine-tuning pLMs for specific viral contexts. 7/ The authors propose a roadmap for developing virus-specific language models, potentially incorporating genomic context and neighborhood information to enhance annotations for understudied viral protein families. @microbegrrl @zflam94 @ASMicrobiology 📜Paper: doi.org/10.1128/mbio.0…













Open search for 2x Asst Prof in my department @SBUPharm @StonyBrookMed @stonybrooku! Please spread the word. Apply here: apply.interfolio.com/150220



