ESM4SL: Protein Language Models for Cancer Cell Line-specific Synthetic Lethality Prediction.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Synthetic lethality (SL) gene pairs are crucial for targeted cancer therapy.
• Identifying SLs is challenging due to data scarcity and poor generalization in existing machine learning models.
• Predicting cell line-specific SLs remains a significant hurdle.

Purpose of the Study:

• To develop a novel approach, ESM4SL, for predicting cell line-specific synthetic lethality.
• To leverage protein language models for enhanced generalization in SL prediction.
• To address data scarcity and weak generalization issues in current SL prediction methods.

Main Methods:

• Utilized the pre-trained protein language model ESM-2 to extract protein sequence and evolutionary information.
• Integrated protein representations using self-attention and cross-attention mechanisms.
• Developed ESM4SL for predicting synthetic lethality gene pairs.

Main Results:

• ESM4SL achieved state-of-the-art performance in predicting cell line-specific synthetic lethality.
• Demonstrated superior generalization ability across multiple scenarios and cell lines.
• Successfully applied protein language models as a core component for SL prediction.

Conclusions:

• ESM4SL offers an effective solution for identifying synthetic lethality gene pairs.
• The approach enhances the prediction of cell line-specific SLs, advancing targeted cancer therapy.
• This work pioneers the use of protein language models in synthetic lethality prediction.