Inter-chain contact prediction based on Mamba and triangular geometric constraints.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in protein science

Background:

• Predicting inter-chain residue-residue contacts is crucial for modeling long protein sequences.
• Existing methods using residual convolutional networks struggle with long-range interactions.

Purpose of the Study:

• To develop an effective framework for protein inter-chain residue-residue contact prediction.
• To address the limitations of current models in capturing long-range dependencies.

Main Methods:

• Proposing MaMCon, a novel framework integrating Mamba (a state space model) and triangular geometric constraints.
• Utilizing Mamba to capture long-range residue interactions.
• Employing triangular geometric constraints to refine paired residue representations.

Main Results:

• MaMCon demonstrates effectiveness in protein inter-chain residue-residue contact prediction.
• Experiments on TEST300, TEST218, and CASP-CAPRI benchmark datasets validate the approach.
• The proposed method shows improved performance compared to existing techniques.

Conclusions:

• MaMCon offers a promising solution for accurate protein contact prediction, especially for long sequences.
• The integration of Mamba and geometric constraints enhances the modeling of complex inter-chain interactions.
• The framework provides a valuable tool for structural bioinformatics research.