Statistically based reduced representation of amino acid side chains

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Protein structure representation often relies on detailed all-atom models and rotamer libraries.
• Predicting side-chain conformations can be computationally intensive and complex.
• Reducing model complexity is crucial for large-scale protein analysis and prediction.

Purpose of the Study:

• To develop a general parametrization for predicting representative side-chain atom positions directly from backbone coordinates.
• To offer a simplified yet accurate method for representing protein side chains.
• To explore applications in areas like protein charge density prediction.

Main Methods:

• Utilized a large dataset of high-resolution protein crystal structures to derive preferred conformations for terminal atoms.
• Developed probabilistic approaches, including Monte Carlo methods, to predict terminal atom locations based on backbone coordinates.
• Validated the prediction method on a separate set of protein structures.

Main Results:

• Identified 1 to 7 preferred conformations for terminal atoms of non-glycine residues.
• Achieved an average root mean-square deviation (RMSD) of approximately 3 Å for terminal atom prediction using basic probabilistic methods.
• Improved prediction accuracy to an average RMSD of 1.74 Å by incorporating conditional probabilities based on side-chain chi(1) rotamer.

Conclusions:

• The proposed method provides a viable and accurate alternative to all-atom rotamers for simplified protein representations.
• This approach effectively reduces model complexity and data handling requirements.
• The method shows strong potential for applications such as predicting protein charge density, particularly on protein surfaces.