#Improving protein language models with minimal wet-lab data #efficiency

The content discusses an efficient solution for the fitness ranking problem using LoRA to fine-tune PLMs with a small number of task-specific parameters. It also covers the use of ListMLE for learning to rank fitness and meta-learning on auxiliary tasks. The content explains the process of selecting mutants for wet-lab experiments on Phi29 DNA polymerase, including plasmid construction, protein expression, purification, and melting temperature assessment. Additionally, it outlines the early stopping strategy and benchmark datasets used in the study. The methodology for training models, such as ESM-1v, ESM-2, and SaProt, is described, along with the protocol for wet-lab experiments. The study aims to predict the effects of mutations on protein function and improve the thermal stability of Phi29 DNA polymerase through a combination of computational and experimental approaches.

Source link

Source link: https://www.nature.com/articles/s41467-024-49798-6