Genomic language model for function prediction of microbial genes – Danielle Miller (TAU)

Title:  Genomic language model for function prediction of microbial genes

Abstract: Revealing the function of uncharacterized genes is a fundamental challenge in an era of ever-increasing volumes of sequencing data. Here, we present a concept for tackling this challenge using deep learning methodologies adopted from natural language processing (NLP). We repurpose NLP algorithms to model “gene semantics” based on a biological corpus of more than 360 million microbial genes within their genomic context. We use the language models to predict functional categories for thousand of genes. We then systematically evaluate the “discovery potential” of different functional categories, pinpointing those with the most genes yet to be characterized. Finally, we demonstrate our method’s ability to discover systems associated with microbial interaction and defense. Our results highlight that combining microbial genomics and language models is a promising avenue for revealing gene functions in microbes.

In this talk, I will walk through the data pipelines and methodology of the approach, demonstrate some of our biological findings and discuss future plans.

Link to the paper: https://www.nature.com/articles/s41467-022-33397-4

About me: Currently a PhD student for bioinformatics jointly supervised by David Burstein & Adi Stern at TAU, working on language models and sequence-based methods for microbial genomics. I have a master in bioinformatics in the field of information theory and its possible applications to viral genomics. My BSc is in Computer Science & Biology from TAU.