LoRA-DR-suite: adapted embeddings predict intrinsic and soft disorder from protein sequences

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Intrinsic disorder regions (IDRs) and soft disorder regions (SDRs) are crucial for protein function, molecular interactions, and assembly.
• These flexible protein regions are implicated in various biological processes and diseases.
• Accurate computational prediction of disordered regions is vital for protein research.

Purpose of the Study:

• To develop a computational tool for precise prediction of IDRs and SDRs.
• To create a comprehensive database of soft disorder information.
• To advance the understanding of protein dynamics and interactions.

Main Methods:

• Utilized a simple adapter-based architecture for LoRA-DR-suite.
• Employed protein language model embeddings for protein sequence representation.
• Constructed the SoftDis database using approximately 500,000 PDB chains.

Main Results:

• LoRA-DR-suite enables precise prediction of IDRs and SDRs directly from primary sequence data.
• The SoftDis database provides a unique resource for soft disorder research.
• The LoRA-DR-suite implementation is fast and efficient.

Conclusions:

• LoRA-DR-suite offers a powerful method for predicting protein disordered regions.
• SoftDis facilitates further research into protein dynamics and interactions.
• These resources advance the study of intrinsically and softly disordered proteins.