Real-to-bin conversion for protein residue distances.

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biochemistry

Background:

• Protein Structure Prediction (PSP) is crucial for understanding protein function.
• Machine learning-driven inter-residue distance prediction significantly advances PSP.
• Binned distance probabilities offer advantages over real-valued distances for differentiable objective functions in PSP.

Purpose of the Study:

• To develop a technique for converting real-valued inter-residue distances into distance bin probabilities.
• To enhance existing Protein Structure Prediction methods by leveraging these converted probabilities.
• To improve the accuracy and performance of 3D protein structure modeling.

Main Methods:

• Proposed a novel method to convert real-valued distances into distance bin probabilities.
• Integrated the real-to-bin conversion technique into existing Protein Structure Prediction pipelines.
• Evaluated the method's performance on standard benchmark protein datasets.

Main Results:

• The real-to-bin converted distances improved PSP methods, yielding 4%-16% better RMSD, TM-Score, and GDT values.
• The proposed method, named real to bin (R2B) inter-residue distance predictor, demonstrated superior performance compared to existing approaches.
• Successfully generated more accurate 3D protein structures.

Conclusions:

• Converting real-valued distances to bin probabilities is an effective strategy to boost Protein Structure Prediction accuracy.
• The R2B predictor offers a valuable tool for the structural biology community.
• The developed code is publicly available to facilitate further research and application.