Скачать или смотреть MULTI-evolve: Rapid directed evolution guided by protein language models and epistatic interactions.

Vincent Tran of the Patrick Hsu Lab discusses his team's publication in the journal Science on machine learning-guided engineering of hyperactive multi-mutant proteins. The authors developed MULTI-evolve, an end-to-end framework that trains neural networks on small datasets of ~200 strategic variants to predict which combinations of beneficial mutations will work synergistically. Applied to three proteins, the framework achieved up to 256-fold improvement in activity while compressing what traditionally takes 5-10 iterative experimental rounds into weeks.

Citation:
Tran, V.Q., Nemeth, M., Bartie, L.J., Chandrasekaran, S.S., Fanton, A., Moon, H.C., Hie, B.L., Konermann, S., & Hsu, P.D. (2026). Rapid directed evolution guided by protein language models and epistatic interactions. Science. https://doi.org/10.1126/science.aea1820

GitHub Link: https://github.com/ArcInstitute/MULTI...

0:00 Introduction
0:11 Enhancing protein function
0:40 Traditional directed evolution
1:08 Machine learning-guided directed evolution (MLDE)
2:05 Key components for rapid protein evolution
2:27 Efficient discovery of function-enhancing mutations
3:23 MULTI-evolve models enable data-efficient extrapolation
4:14 Reliable construction of multi-mutant genes
4:36 MULTI-evolve framework
5:24 Developing a rapid protein evolution approach
6:09 MULTI-evolve applied to three distinct proteins
6:38 Acknowledgements