FINDSITE(X): a structure-based, small molecule virtual screening approach with application to all identified human...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Drug Discovery

Background:

• Traditional protein binding site prediction relies on solved holo protein structures.
• Limitations exist in obtaining sufficient homologous protein structures for analysis.
• Accurate prediction of protein structures and ligand binding is crucial for understanding function.

Purpose of the Study:

• To develop FINDSITE(X), an improved algorithm for protein binding site inference and virtual ligand screening.
• To overcome the dependency on solved holo protein structures by utilizing predicted structures.
• To enhance the identification of functional relationships between proteins and their ligands.

Main Methods:

• Development of TASSER(VMT)-lite for fast and accurate template-based protein structural modeling.
• Hybrid approach combining structure alignments and evolutionary similarity scores for functional relationship detection.
• Application of FINDSITE(X) to screen human G-protein coupled receptors (GPCRs) against large ligand databases.

Main Results:

• FINDSITE(X) demonstrates significantly improved performance over FINDSITE in virtual screening enrichment factors (EF(0.01) of 22.7 vs. 7.1, and 41.4 when excluding native ligands).
• TASSER(VMT)-lite provides accurate predicted protein structures comparable to top CASP9 servers.
• Analysis of off-target interactions for a set of 168 human GPCRs was performed.

Conclusions:

• FINDSITE(X) offers a powerful and accurate method for protein binding site prediction and virtual screening, especially when experimental structures are unavailable.
• The integration of predicted structures and advanced similarity scoring expands the applicability of computational drug discovery.
• The developed tools and data for human GPCRs are publicly accessible, facilitating further research.