Protein sidechain conformer prediction: a test of the energy function

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Science

Background:

• Homology modeling is crucial for predicting protein structures from sequences.
• Accurate sidechain placement and loop modeling remain significant challenges in homology modeling.
• Current sidechain prediction methods achieve approximately 70% accuracy, with potential limitations in energy functions.

Purpose of the Study:

• To evaluate the accuracy of the CHARMM potential energy function for predicting protein sidechain orientations.
• To identify limitations in current energy-based sidechain prediction methods.
• To establish a benchmark for evaluating future sidechain prediction algorithms.

Main Methods:

• Calculated minimum energy positions for individual sidechains in ten proteins.
• Assessed sidechain prediction accuracy for chi1 and chi2 rotations.
• Evaluated the impact of crystal waters on prediction accuracy.
• Correlated prediction accuracy with energy gaps between primary and secondary minima.

Main Results:

• CHARMM function achieved 77.4% (chi1) and 89.5% (chi2) accuracy without waters.
• Accuracy improved to 86.5% (chi1) and 94.9% (chi2) in the presence of crystal waters.
• Prediction errors often corresponded to alternative local energy minima.
• Accuracy correlated positively with the energy gap between the global and local minima.

Conclusions:

• The CHARMM potential energy function is not the primary source of errors in current sidechain prediction.
• The developed test serves as a necessary condition for validating energy-based and statistical sidechain prediction methods.
• Further improvements in sidechain prediction accuracy may require addressing factors beyond the potential energy function.