FingerprintContacts: Predicting Alternative Conformations of Proteins from Coevolution

Area of Science:

• * Computational biology and structural bioinformatics.
• * Machine learning applications in molecular biology.

Background:

• * Proteins are dynamic molecules with diverse functions, adopting various 3D structures.
• * Predicting protein structure from sequence is crucial for understanding function.
• * Current residue-residue contact prediction methods struggle to capture multiple protein conformations.

Purpose of the Study:

• * To introduce FingerprintContacts, a novel machine learning method.
• * To enhance the prediction of protein structure and conformational diversity from residue-residue contacts.
• * To improve the understanding of protein functional mechanisms.

Main Methods:

• * Development of a machine learning algorithm, FingerprintContacts.
• * Leveraging specific features of residue-residue contacts for structural prediction.
• * Applying the method to eight ligand-binding proteins with conformational variability.

Main Results:

• * FingerprintContacts effectively utilizes residue-residue contacts for structural prediction.
• * The method identifies conformation-specific contacts.
• * Small clusters of contacts in dynamically fluctuating regions were identified.

Conclusions:

• * FingerprintContacts extends the utility of residue-residue contacts for protein structure prediction.
• * The method aids in understanding protein dynamics and functional mechanisms.
• * FingerprintContacts serves as a foundational tool for structural biology research.