Check out our new perspective "Illuminating the universe of enzyme catalysis in the era of artificial intelligence" now out in @CellSystemsCP !. We discuss a vision and path forward for genetically encoding almost all chemistry, powered by new AI tools:.

RT @AnimaAnandkumar: Very pleased to see our AI model GenSLM designing novel and versatile enzymes in a challenging setting in @francesarno….

RT @AlexanderTong7: Thrilled to announce I'm starting as a Principal Investigator at #Aithyra in Vienna! We'll be developing generative mod….

RT @KevinKaichuang: A compelling review of how ML/AI could help in the quest to find an enzyme for every reaction. @jsunn_y @francescazfl….

RT @francesarnold: I think we will soon have AI tools to genetically encode--i.e., make #enzymes for--many useful chemical transformations.….

This was fun to write, with special thanks to my co-authors @francescazfl, @Yueming_Long_, and @francesarnold.

RT @ginaelnesr: The MLSB workshop will be in San Diego, CA (co-located with NeurIPS) this year for its 6th edition in December 🧬🔬. Stay tun….

RT @yisongyue: @jsunn_y, @WendaChu32619 & team doing some awesome work in helping us better understand wetlab-in-the-loop guided diffusion….

RT @pranamanam: I absolutely love this paper!! 🌟 Beautiful work by @jsunn_y @yisongyue and team to show that discrete diffusion models can….

RT @BiologyAIDaily: Steering Generative Models with Experimental Data for Protein Fitness Optimization. １．This paper introduces SGPO (Steer….

RT @KevinKaichuang: Steered generation for protein optimization: On datasets with ~10^2 measurements, steering a discrete diffusion model o….

Thank you to my co-lead @WendaChu32619 and my amazing collaborators Daniel Khalil, Raul Astudillo, Bruce Wittmann, @francesarnold, and @yisongyue! (🧵5/5).

On the TrpB and CreiLOV datasets, we found that these strategies are highly effective, and guiding diffusion models shows the strongest performance with low computational costs. We expect that these methods can be incorporated into real-world wet-lab workflows! (🧵4/5).

Specifically, we comprehensively compared methods for steering generative models, such as classifier guidance with discrete diffusion models and reinforcement learning with language models, for optimization when only hundreds of real-world fitness labels are available. (🧵3/5)

Protein fitness optimization entails finding a sequence that maximizes desired properties in a massive design space. In this project, we focused on generative methods that utilize both unlabeled data on natural sequences and assay-labeled fitness measurements. (🧵2/5)

I'm excited to share our new preprint "Steering Generative Models with Experimental Data for Protein Fitness Optimization" (🧵1/5)!. Paper: Code:

I’ll be at #ICLR2025 in Singapore this week! I’ll also be presenting some new work at the GEM workshop on Sun, April 27. Please reach out if you want to link up!.

Happy to share that our work on Active Learning-Assisted Directed Evolution is now published in @NatureComms! We show that it's an effective and broadly applicable method to accelerate protein engineering with machine learning. Paper:

Come visit our CARE benchmarks poster today at 11-2pm! (West Ballroom #5205).

I’m at NeurIPS this week presenting two of my recent projects!. Enzyme function (CARE) benchmarks: Friday 11-2pm West Ballroom #5205 @Caltech . Conditional generation from PLMs (ProCALM): Sunday at MLSB Workshop @ProfluentBio . Please come say hi!.