Linker-Tuning: Optimizing Continuous Prompts for Heterodimeric Protein Prediction

• DOI: 10.48550/arxiv.2312.01186
• Publication Date: 2023-01-01
• Authors: Shuxian Zou; Hui Li; Shentong Mo; Xingyi Cheng; Eric P. Xing; Le Song
• Keywords: Linker; Sequence (biology)

Predicting the structure of interacting chains is crucial for understanding biological systems and developing new drugs. Large-scale pre-trained Protein Language Models (PLMs), such as ESM2, have shown impressive abilities in extracting biologically meaningful representations protein prediction. In this paper, we show that ESMFold, which has been successful computing accurate atomic structures single-chain proteins, can be adapted to predict heterodimer a lightweight manner. We propose Linker-tuning, learns continuous prompt connect two dimer before running it single sequence ESMFold. Experiment results our method successfully predicts 56.98% interfaces on i.i.d. test set, with an absolute improvement +12.79% over ESMFold-Linker baseline. Furthermore, model generalize well out-of-distribution (OOD) set HeteroTest2 antibody sets Fab Fv while being $9\times$ faster than AF-Multimer.