Text mining improves prediction of protein functional sites

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Predicting protein functional sites is crucial for understanding biological processes.
• Existing methods often lack comprehensive integration of structural and textual data.

Purpose of the Study:

• To introduce LEAP-FS (Literature Enhanced Automated Prediction of Functional Sites), an integrated approach for protein functional site prediction.
• To evaluate the performance of structure-based (Dynamics Perturbation Analysis) and text-mining methods, individually and combined.

Main Methods:

• Protein structure analysis using Dynamics Perturbation Analysis (DPA) to identify vibrationally perturbed regions.
• Text mining to extract residue mentions from scientific literature, inferring functional importance.
• Assessing prediction accuracy against known small-molecule binding and catalytic sites in ~100,000 protein structures.

Main Results:

• DPA successfully identified known functional sites, prioritizing biologically relevant binding sites.
• Residues mentioned in literature were significantly more likely (approx. 6x) to be functional sites.
• The combined LEAP-FS approach showed improved prediction accuracy when structure-based and text-based predictions converged.

Conclusions:

• Both DPA and text mining independently offer valuable high-throughput functional site predictions.
• Integrating Dynamics Perturbation Analysis and literature text mining via LEAP-FS enhances the quality and reliability of protein functional site predictions.
• LEAP-FS identified novel functional site residues not present in current curated databases.