AI Seminar: Inferring selection effects in SARS-CoV-2 with Bayesian Viral Allele Selection


Join us for a talk by Martin Jankowiak. Everybody is welcome to attend.


Inferring selection effects in SARS-CoV-2 with Bayesian Viral Allele Selection


Over 15 million SARS-CoV-2 viral genomes have been sequenced since the beginning of the pandemic, giving an unprecedented view into viral evolution. Methods to identify emerging variants of epidemiological significance and characterize mutational determinants of enhanced viral fitness are important for public health, but the scale and complexity of genomic surveillance data make rigorous analysis challenging. To meet this challenge we develop Bayesian Viral Allele Selection (BVAS), a principled probabilistic model that combines an elegant diffusion-based likelihood with Bayesian Variable Selection.

By providing a genome-wide view of the evolution of SARS-CoV-2, our model-based approach complements more targeted experimental approaches for elucidating the functional consequences of different viral mutations. We argue that running BVAS periodically as part of a real-time genomic surveillance program could provide valuable information for public health authorities about new lineages as they emerge.


Martin Jankowiak is a machine learning scientist at Generate Biomedicines and a co-creator of the Pyro probabilistic programming framework. Martin has a PhD in theoretical physics from Stanford and works on a wide range of topics across probabilistic machine learning, bayesian inference, and applications in computational biology.

Relevant links: