Multiple search methods for similarity-based virtual screening: analysis of search overlap and precision

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Data fusion methods are common in virtual screening.
• These methods assume molecules retrieved more often across searches are more likely active.
• This study evaluates the validity of this core assumption.

Purpose of the Study:

• To rigorously test the assumption underlying data fusion in virtual screening.
• To determine if increased retrieval frequency correlates with higher molecular activity.
• To provide a quantitative analysis of search consolidation in drug discovery.

Main Methods:

• Conducted sets of 25 similarity searches.
• Employed two fusion strategies: varying similarity measures with a single reference structure (similarity fusion) and varying reference structures with a single measure (group fusion).
• Analyzed the number of unique molecules retrieved and their activity fractions.

Main Results:

• A large number of unique molecules were retrieved by single searches.
• The count of unique molecules rapidly decreased as more searches were aggregated.
• This decrease correlated with a sharp increase in the proportion of active molecules found.
• A power-law relationship was observed between unique molecules and the number of searches.

Conclusions:

• Multiple similarity searches effectively increase the precision of virtual screening.
• The findings validate the use of data fusion techniques in computational drug discovery.
• Consolidating search results is a viable strategy for improving hit rates.