Highly accurate protein structure prediction with AlphaFold

John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Figurnov, Olaf Ronneberger

Nature, 2021

"The speaker cites AlphaFold as a prime example of a predictive protein model that is powered by evolutionary signals, specifically using a multiple sequence alignment (MSA) as input, which is crucial for its high performance."

State-of-the-art estimation of protein model accuracy using AlphaFold

James P Roney, Sergey Ovchinnikov

bioRxiv, 2022

"This paper is cited to support the claim that the initial structural guess provided by co-variation in the multiple sequence alignment is crucial for AlphaFold's performance."

The topology and geometry of neural representations

Baihan Lin, Nikolaus Kriegeskorte

arXiv [q-bio.NC], 2023

"This paper is cited as providing analysis suggesting that protein language models like ESM-2, which use single sequences, still show similar failure modes to statistical models of sequence evolution, implying they also learn from evolutionary signals."

A statistical-based method for the construction and analysis of gene network: application to bacteria

Zhiyuan Zhang, Guozhong Chen, Erguang Li

bioRxiv, 2024

"This paper is cited alongside Lin et al. 2023 to support the idea that protein language models learn from evolutionary signals present in sequence data."

Systems and algorithms for convolutional multi-hybrid language models at scale

Jerome Ku, Eric Nguyen, David W Romero, Garyk Brixi, Brandon Yang, Anton Vorontsov

ArXiv, 2025

"The speaker presents this as a companion preprint that provides the details of the algorithm and systems work (StripedHyena 2 architecture) that enabled the efficient training of their Evo 2 model."

Anthropic Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet

"This paper from Anthropic is cited as an example of mechanistic interpretability finding meaningful features in large language models, specifically a feature that identifies bugs in computer code, which serves as an analogy for their own work finding features for "bugs" (frameshifts) in genetic code."

Genome modeling and design across all domains of life with Evo 2

Garyk Brixi, Matthew G Durrant, Jerome Ku, Michael Poli, Greg Brockman, Daniel Chang

bioRxiv, 2025

"This is the main preprint describing the work presented in the talk, which the speaker encourages the audience to read."