Three-body interactions improve contact prediction within direct-coupling analysis.

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein science

Background:

• Predicting protein residue contacts from sequence data is crucial for determining protein structure.
• Current methods often rely on statistical or information-theoretic approaches using multiple sequence alignments.
• The accuracy of existing two-body interaction models within the Potts model is limited.

Purpose of the Study:

• To enhance the accuracy of protein residue contact prediction.
• To generalize existing Potts models by incorporating three-body interactions.

Main Methods:

• Developed a generalized Potts model Hamiltonian that includes an additional three-body interaction term.
• Derived a fast mean-field approximation for inferring three-body couplings.
• Evaluated the model's performance on predicting residue contacts.

Main Results:

• The proposed model with three-body interactions demonstrates higher accuracy in predicting residue contacts compared to models using only two-body interactions.
• The mean-field approximation for three-body coupling inference is computationally efficient on modern hardware.

Conclusions:

• Incorporating three-body interactions in Potts models significantly improves the prediction of protein residue contacts.
• The developed mean-field approximation provides a computationally feasible approach for applying these enhanced models.
• This work advances computational structure prediction by offering a more accurate method for contact prediction from sequence information.