Linear programming optimization and a double statistical filter for protein threading protocols.

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Protein threading is crucial for predicting protein structures.
• Existing scoring functions, like pairwise and profile models, have limitations in accuracy and computational cost.
• Optimizing these functions is key to advancing structural prediction.

Purpose of the Study:

• To design optimal scoring functions for protein threading using linear programming (LP).
• To develop a new profile model with prediction capacity comparable to contact potentials.
• To present an efficient and accurate method for protein structure prediction.

Main Methods:

• Utilized linear programming (LP) to design and optimize scoring functions.
• Compared the prediction capacity of pairwise and profile models for gapless and gapped threading.
• Developed and tested a novel profile model and a protocol for optimizing gap parameters.

Main Results:

• Pair potentials showed higher prediction capacity in gapless threading compared to profile energies.
• The new profile model demonstrated prediction capacity comparable to contact potentials.
• The proposed protocol for gap parameter optimization using LP was effective.
• A statistical test combining local and global Z-scores was used to filter false positives.

Conclusions:

• The new profile model offers an efficient alternative to pair energies for protein threading.
• The developed methods maintain comparable accuracy to existing approaches.
• The study provides a valuable tool and server for protein structure prediction.