Which Mutations Make a Virus Spread? Reading 6.4 Million Genomes
Dr. Jacob E. Lemieux
Science · 2022Paper discussed
An interpretable Bayesian model that turns millions of viral genomes into a fitness estimate for each mutation.
Speaker
Infectious disease physician-scientist, Massachusetts General Hospital
Dr. Jacob E. Lemieux is an infectious disease physician-scientist at Massachusetts General Hospital and Harvard Medical School whose group studies the microbial genetics of infectious diseases, including COVID-19. He is a corresponding author of the PyR0 study, published in Science in 2022 with lead author Fritz Obermeyer and colleagues at the Broad Institute of MIT and Harvard.
Jacob E. Lemieux is an infectious disease physician-scientist. The 2022 Science paper discussed in his episode lists his affiliations as the Broad Institute of MIT and Harvard and the Division of Infectious Diseases at Massachusetts General Hospital, and names him as a corresponding author with equal contribution. His group studies the microbial genetics of infectious diseases, including COVID-19.
The paper is Analysis of 6.4 million SARS-CoV-2 genomes identifies mutations associated with fitness, published in Science in 2022, with Fritz Obermeyer as first author. It introduces PyR0, which its abstract calls a hierarchical Bayesian multinomial logistic regression model. PyR0 estimates how common each viral lineage is across regions, flags lineages that are growing, and identifies mutations linked to fitness. Applied to the publicly available genomes, it found many fitness-increasing substitutions, including previously identified spike mutations and many in the nucleocapsid and nonstructural proteins. The authors report that the model can forecast the growth of new lineages from their mutation profiles, rank lineage fitness as new sequences arrive, and prioritize mutations for functional study.
In Episode 002 Lemieux walks through how the model is formulated and written in the Pyro probabilistic programming language, what its results say about mutation-specific fitness, and where it falls short, with chapters on limitations and on comparing methods. On Interpretable the model matters because it attributes growth to individual mutations, with uncertainty, so a forecast arrives with a reason that virologists and public health teams can check in the laboratory. The episode is filed under AI in Health, Public Health, Infectious Disease, Genomics, and Interpretable AI.
Dr. Jacob E. Lemieux
Science · 2022Paper discussed
An interpretable Bayesian model that turns millions of viral genomes into a fitness estimate for each mutation.