A benchmark of optimally folded protein structures using integer programming and the 3D-HP-SC model.

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• The Protein Structure Prediction (PSP) problem is crucial for understanding protein function.
• Computational studies often rely on synthetic data, limiting real-world applicability.
• Accurate PSP requires robust models and realistic datasets.

Purpose of the Study:

• To establish a benchmark for PSP using real protein sequences.
• To evaluate the 3D-Hydrophobic-Polar with Side-Chains (3D-HP-SC) model.
• To explore optimization methods for predicting protein structures.

Main Methods:

• Utilized 17 real protein sequences from the Protein Data Bank.
• Employed the tri-dimensional Hydrophobic-Polar with Side-Chains (3D-HP-SC) model.
• Solved the protein structure prediction as an optimization problem using Integer Programming.

Main Results:

• Maximized hydrophobic contacts between amino acid side-chains to determine native structure.
• The Integer Programming method provides optimal solutions but exhibits exponential processing time.
• A linear relationship was found between hydrophobic contacts and residue count, enabling upper bound estimation.
• Highest similarity between predicted and biological structures observed at 5.2-8.2 Å distance thresholds.

Conclusions:

• The developed method is a proof-of-concept for small sequences due to computational limitations.
• The created dataset and programs are available to advance PSP research.
• This work highlights the importance of real biological data in protein structure prediction.