Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
This talk from Valence Labs features Mangal Prakash and Artem Moskalev presenting their research on HELM (Hierarchical Encoding for mRNA Language Modeling), a novel approach to modeling messenger RNA sequences. Discover how HELM incorporates the crucial codon-level hierarchical structure of mRNA into language model training, addressing limitations of previous approaches that failed to account for this biological reality. Learn about their innovative pre-training strategy that modulates the loss function based on codon synonymity, resulting in approximately 8% performance improvements over standard language models and existing foundation model baselines across six diverse downstream property prediction tasks and antibody region annotation. The presentation also covers how HELM enhances generative capabilities, producing mRNA sequences that better align with true biological data distributions compared to non-hierarchical approaches. This 59-minute research presentation is part of the AI for drug discovery community hosted on Portal by Valence Labs.
