Information and discrimination in pairwise contact potentials.

Area of Science:

• Bioinformatics
• Structural Biology
• Information Theory

Background:

• Statistical potentials are crucial for protein structure prediction.
• Current methods often rely on empirical data and can be computationally intensive.
• Optimizing these potentials is key to improving prediction accuracy.

Purpose of the Study:

• To develop an efficient, information-theoretic strategy for optimizing statistical potentials.
• To enhance the utilization of structural information from empirical data.
• To improve the effectiveness of contact potentials in protein structure prediction.

Main Methods:

• Established connections between information-theoretic quantities (e.g., Z-score) and contact scores.
• Quantified the fold information content of pairwise residue contacts.
• Evaluated performance using threading tests and analyzed parameter influences.

Main Results:

• Total divergence was identified as the key quantity for fold discrimination.
• Optimized pairwise contacts significantly improve native conformation identification.
• The derived 'information product' metric is as effective as the Z-score but computationally faster.

Conclusions:

• Potentials optimized for mutual information and high score events are superior for fold recognition.
• The 'information product' offers a faster alternative to Z-score for parameter optimization.
• This approach reduces reliance on biophysical formalisms, enabling direct bioinformatic optimization.