Authors: Bowen Jing, Bonnie Berger, Tommi Jaakkola

Published on: February 07, 2024

Impact Score: 8.15

Arxiv code: Arxiv:2402.04845

Summary

• What is new: Development of flow-based generative models, AlphaFlow and ESMFlow, for protein structure sampling, leveraging and fine-tuning existing single-state predictors.
• Why this is important: The need to understand proteins’ dynamic structural ensembles, which is critical for elucidating their biological functions.
• What the research proposes: A flow-based generative modeling approach that repurposes AlphaFold and ESMFold, creating generative models that can accurately capture the conformational landscapes of proteins.
• Results: Superior precision and diversity in modeling protein structures compared to existing methods, with added ability to capture conformational flexibility, positional distributions, and ensemble observables for unseen proteins.

Technical Details

Technological frameworks used: Custom flow matching framework, repurposing AlphaFold and ESMFold

Models used: AlphaFlow and ESMFlow

Data used: PDB, all-atom MD ensembles

Potential Impact

Pharmaceuticals, Biotechnology, Computational biology platforms

Want to implement this idea in a business?

We have generated a startup concept here: FlexiFold.