Influence of protein structure databases on the predictive power of statistical pair potentials

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Developing reliable energy functions is crucial for protein structure verification and theoretical studies.
• Knowledge-based statistical pair potentials offer a low-cost method to model protein structures and solvent effects.

Purpose of the Study:

• To investigate the derivation and characteristics of statistical pair potentials from protein 3D structures.
• To address how database content and form influence the performance of these potentials.

Main Methods:

• Analysis of statistical pair potentials derived from protein three-dimensional structure databases.
• Investigating the impact of database size and residue proximity (spatial cutoff) on potential accuracy.
• Evaluating the influence of secondary structure diversity within the database.

Main Results:

• Statistical pair potentials are dependent on the size of proteins in the derivation database.
• A spatial cutoff of 8 Å for residue pairs reduces this size dependence.
• Potentials reflect the secondary structure composition of the database, impacting fold recognition accuracy.

Conclusions:

• The quality and diversity of the protein structure database are critical for deriving effective statistical potentials.
• Potentials derived from databases with limited structural diversity (e.g., only alpha-proteins) perform best on similar structures.
• Careful database construction is essential to mitigate inherent weaknesses in statistical potentials.