DARMN: Domain-Aware Residual Feature Modulation Network for Multidomain Protein Dynamic Inter-Residue Contact...

Area of Science:

• Computational Biology
• Protein Engineering
• Structural Bioinformatics

Background:

• Multidomain proteins (MDP) regulation is key for function, involving domain interactions.
• Dynamic inter-residue contacts govern conformational transitions but are hard to detect experimentally.
• Current deep learning methods struggle with specialized modeling for MDP dynamics.

Purpose of the Study:

• To develop a deep learning framework for predicting dynamic inter-residue contacts in MDP.
• To improve the understanding of conformational stability and function in MDP.
• To provide a computational tool for protein engineering and mechanistic studies.

Main Methods:

• Developed the Domain-Aware Residual Feature Modulation Network (DARMN) using AlphaFold2 representations.
• Integrated coevolutionary information to capture interdomain contact signals.
• Implemented a domain-aware weighted focal loss for distinguishing intra- and interdomain contacts.

Main Results:

• DARMN effectively predicts dynamic inter-residue contacts, outperforming existing methods.
• The model shows particular strength in identifying long-distance contacts.
• DARMN demonstrates good generalization to novel multidomain proteins.

Conclusions:

• DARMN offers a robust computational approach for predicting dynamic contacts in MDP.
• This framework aids in sequence design for enhanced protein stability.
• DARMN facilitates mechanistic investigations of multidomain protein function.