Factors influencing the ability of knowledge-based potentials to identify native sequence-structure matches

Area of Science:

• Computational biology
• Structural bioinformatics
• Protein structure prediction

Background:

• Protein structure prediction is crucial for understanding protein function.
• Developing accurate potentials is key to reliable computational prediction.
• Existing methods face challenges in accurately determining full three-dimensional folds.

Purpose of the Study:

• To derive and assess novel potentials for protein structure prediction.
• To evaluate the performance of different potential types in recognizing native protein folds.
• To investigate the impact of various interaction descriptions on prediction accuracy.

Main Methods:

• Statistical analysis of known protein structures to derive potentials.
• Formulation of potentials for backbone dihedral angles, residue interactions, and solvation.
• Rigorous screening test involving threading sequences through structural motifs.
• Assessment of potentials based on side-chain centroid, C-alpha, and C-beta distances.

Main Results:

• Potentials based on side-chain centroid distances outperform those using C-alpha or C-beta distances.
• Combining different types of potentials enhances recognition accuracy.
• Backbone dihedral angle potentials successfully recognized 68 out of 74 protein chains.
• Some potentials achieved perfect native fold recognition for all tested proteins.

Conclusions:

• Statistical potentials derived from protein structure data are effective for native fold recognition.
• Side-chain centroid interactions offer a robust basis for residue interaction potentials.
• The effectiveness of these potentials raises questions about the stringency of current screening tests.
• Further refinement of potentials and testing methodologies is warranted.