TEMC-Cas: Accurate Cas Protein Classification via Combined Contrastive Learning and Protein Language Models.

Area of Science:

• Biotechnology
• Genomics
• Bioinformatics

Background:

• Accurate classification of Cas proteins is essential for understanding CRISPR-Cas systems and developing genome-editing technologies.
• Traditional methods like BLAST and HMMs struggle with distant homology and closely related subtypes.

Purpose of the Study:

• To present TEMC-Cas, a novel deep learning framework for precise Cas protein classification.
• To leverage evolutionary-scale modeling and contrastive learning for enhanced classification accuracy.

Main Methods:

• Utilized a finely tuned ESM protein language model combined with contrastive learning.
• Incorporated LoRA for efficient parameter adaptation and weighted loss functions to handle class imbalance.
• Employed evolutionary-scale modeling to capture distant homology and contrastive learning to differentiate subtypes.

Main Results:

• TEMC-Cas demonstrated superior performance in classifying Cas1-Cas13 families and 17 Cas12 subtypes.
• The framework showed particular strength in identifying remote homology between Cas proteins.
• Achieved high accuracy in distinguishing closely related Cas protein subtypes.

Conclusions:

• TEMC-Cas offers a robust and accurate method for Cas protein classification.
• This framework enhances the discovery of CRISPR systems and expands genome engineering capabilities.
• The tool is freely accessible, promoting further research and application in the field.