Our newest work is out! Single Layers of Attention Suffice to Predict Protein Contacts This was a wonderful collaboration with Nick Bhattacharya, @roshan_m_rao Justas Dauparas David Baker (Institute for Protein Design) Peter Koo Yun S. Song Sergey Ovchinnikov biorxiv.org/content/10.110… 🧵 1/n

We have ≥$10k to support talented 14-18 year olds whose studies were interrupted by war in Ukraine. We especially would like to hear from IMO, EGMO, MEMO, IOI, EGOI, IPhO, IChO contestants. If you're one or know one, here's the form to apply: ferenchuszar1.typeform.com/ukrainefund-eng

A Stanford University professor just threatened me with police. After BBQ Becky, Permit Patty, Golfcart Gail, and all the memes, we now have Retweet Rachel. Public advisory: don't call the cops on black people for no reason. Black people disagreeing with you on Twitter is not a crime.

Predicting the effects of missense variants is a central problem in human genome interpretation. We are thrilled to share our preprint on using cross-protein transfer (CPT) learning to improve zero-shot prediction of disease variant effects: doi.org/10.1101/2022.1… (1/8)

Interested in a fast, accurate method for estimating the parameters of a complex phylogenetic model of molecular evolution (e.g. a general rate matrix describing the co-evolution of protein contact sites in 3D)? If so, please check out our preprint: doi.org/10.1101/2022.1… (1/10)

Very glad to see our CherryML method published online in Nature Methods. The published version has updated and expanded results, so please check it out. doi.org/10.1038/s41592… (1/3)

I made the image below for our CherryML paper. doi.org/10.1038/s41592… Unfortunately, it did not get selected as cover art, so I am sharing it here. I had fun making it, especially since I rarely have the opportunity to draw or paint these days.

After lengthy anticipation, we finally got to compare PrimateAI-3D with our Cross-Protein Transfer (CPT) model doi.org/10.1186/s13059… Below is a summary of our findings, including a comparison of ESM-1b and ESM-1v protein language models. 1/8

We recently posted a preprint describing GPN-MSA, a DNA language model that leverages whole-genome alignments across multiple species while taking only a few hours to train. This thread summarizes its performance on the human genome. doi.org/10.1101/2023.1… 1/12

Incredible Result for the University of Buenos Aires!!! They definitely deserve to get awarded a bronze medal per the ICPC World Final Rules (just 15 minutes of penalty more than the team placed #12). It would also be the first medal for a Latin American team in over 20 years!

An independent benchmarking study comparing 84 variant effect predictors ranked our method, CPT, at the top for DMS data, beating AlphaMissense. Many thanks to Joe Marsh for conducting this study! Marsh lab paper: doi.org/10.1101/2024.0… Our CPT paper: doi.org/10.1186/s13059…

Antibodies are highly diverse, but most possible sequences are unstable or polyreactive. In this work led by Milind Jagota, we propose a new source of data for modeling constraints from these properties. Our models show clear improvements in predicting antibody dysfunction.(1/n)

Large protein language models can learn complex epistatic interactions, but how much does that help with predicting variant effects? In this NeurIPS article, we show that classical independent-sites phylogenetic models can outperform pLMs on this task. 1/2 openreview.net/forum?id=H7mEN…

Happy New Year! Our GPN-MSA paper is finally published, under a slightly different title from the preprint. Please check it out and share it with your colleagues. doi.org/10.1038/s41587… 1/4

How can one efficiently simulate phylodynamics for populations with billions of individuals, as is typical in many applications, e.g., viral evolution and cancer genomics? In this work with Michael Celentano, W. DeWitt, & S. Prillo, we provide a solution. doi.org/10.1073/pnas.2… 1/n