IGI Seminar Series: Deep sequencing and deep learning for mRNA design

Summary

Revolutionary advances in synthetic mRNA vaccines and therapeutics have exposed a need to understand the rules of mRNA design. Although there is a known strong preference for some codons over others, synonymous mutations that change the codon but not the amino acid are widely regarded as inconsequential. Nonetheless, our work has recently shown that codon choice drives a quantitative difference in protein output: in general, the slower the decoding of a gene, the less protein is made. This overall preference for faster codons is countered by idiosyncratic and hard-to-predict cases where fast translation can disrupt function, eg, by interfering with co-translational protein folding. Using a combination of massively parallel experiments and large language models, we are building models of the constraints on mRNA sequences that will enable precise design of synthetic mRNAs optimized for function in a diverse range of conditions.

Speaker

Liana Lareau — An assistant professor in the Department of Bioengineering at Berkeley, where she use quantitative methods to understand regulation of gene expression. Her work combines high-throughput experimental methods with computational analysis.

She was a postdoctoral fellow in Pat Brown‘s research group in the Biochemistry department at Stanford University, supported by a fellowship from the Damon Runyon Cancer Research Foundation. She used bioinformatics and experimental methods to learn how ribosomes move along transcripts. After Pat’s lab closed down, she spent a year at UC Santa Cruz working on a genome evolution project with Ed Green in the Paleogenomics Lab, before moving back to Berkeley and starting her own research group as a QB3 Distinguished Fellow.

She did my PhD in the Molecular & Cell Biology department at UC Berkeley with Steven Brenner, and her undergrad degrees were in math and biology at MIT.