DNCON2_Inter: predicting interchain contacts for homodimeric and homomultimeric protein complexes using multiple...

Area of Science:

• Computational biology
• Structural bioinformatics
• Deep learning applications in protein science

Background:

• Predicting interchain protein contacts is crucial for understanding protein complex structures.
• Existing deep learning methods often require multiple sequence alignments (MSAs) of protein pairs, which are difficult to obtain for many protein complexes.
• Homomultimeric proteins offer an alternative source of co-evolutionary information within monomeric MSAs.

Purpose of the Study:

• To develop a deep learning method (DNCON2_Inter) for predicting interchain residue-residue contacts in homomultimeric proteins using only monomeric MSAs.
• To evaluate the accuracy of DNCON2_Inter in predicting interchain contacts.
• To demonstrate the utility of predicted contacts for reconstructing protein complex structures using a companion tool (Con_Complex).

Main Methods:

• Applied DNCON2, a deep learning contact predictor, to monomeric MSAs and co-evolutionary features.
• Developed a discrimination strategy to distinguish interchain from intrachain contacts based on monomer tertiary structure.
• Integrated predicted contacts into Con_Complex for *de novo* complex structure reconstruction.

Main Results:

• DNCON2_Inter achieved significant precision in predicting interchain contacts for homodimers (22.9% Top-L/10) and higher-order homomultimers (17.0%).
• In cases of high interchain contact density, DNCON2_Inter demonstrated up to 100% precision.
• Con_Complex successfully reconstructed some homomeric protein complex structures using the predicted contacts.

Conclusions:

• Monomeric MSAs contain sufficient co-evolutionary signals for predicting interchain contacts in homomultimeric proteins.
• DNCON2_Inter provides a viable alternative for predicting interchain contacts when paired MSAs are unavailable.
• The combination of DNCON2_Inter and Con_Complex enables accurate structure prediction of homomeric protein complexes.