Protein structure prediction provides comparable performance to crystallographic structures in docking-based virtual...

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Structure-based virtual screening typically requires experimental protein structures or high-resolution models from strong templates.
• The efficacy of protein structure prediction in virtual screening for targets with weak homology is largely unexplored.

Purpose of the Study:

• To evaluate the performance of I-TASSER-generated protein models in structure-based virtual screening when only weak homology templates are available.
• To assess the potential of advanced protein modeling in drug discovery for targets lacking experimental structures.

Main Methods:

• Utilized the DUD (Dictionary of Unique Decoys) database for virtual screening.
• Employed the I-TASSER protein structure prediction tool with templates exhibiting low sequence identity (<30%).
• Compared the enrichment rates of I-TASSER models against experimental crystal structures.

Main Results:

• I-TASSER models achieved comparable enrichment rates to experimental structures in most cases, even with weak homology templates.
• For 65% of targets, I-TASSER models reached 70% of the virtual screening performance of holo-crystal structures.
• Model accuracy in predicting protein fold and binding pocket structures correlated with virtual screening success.

Conclusions:

• I-TASSER-based structure modeling is a viable approach for virtual screening, particularly for proteins without experimental structures.
• Combining structure-based docking with advanced protein modeling enhances large-scale drug screening and discovery efforts.
• This strategy broadens the scope of targets amenable to structure-based drug design.