Local neighborhood behavior in a combinatorial library context.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• The molecular similarity principle is crucial in drug discovery.
• The Neighborhood Behavior (NB) optimality criterion aids in selecting optimal descriptor spaces and similarity metrics.
• Previous work focused on global NB criteria for descriptor space selection.

Purpose of the Study:

• To introduce and evaluate a local Neighborhood Behavior (NB) score for assessing similarity-based virtual screening performance.
• To analyze the performance of various descriptor/metric combinations using individual bioactive compounds as queries.
• To compare the effectiveness of 2D and 3D pharmacophore-based descriptors in local virtual screening.

Main Methods:

• Systematic studies of local NB were performed on a large combinatorial library of compounds.
• Compounds had reported IC50 values for five proteases.
• Over 140 descriptor/metric combinations, including fragment- and pharmacophore-based descriptors with different similarity metrics, were evaluated.

Main Results:

• The NB-score's performance was highly dependent on the specific query compound for each descriptor/metric combination.
• On average, 2D pharmacophore-based descriptors demonstrated superior performance compared to their 3D counterparts.
• This suggests that 2D descriptors may be more effective for local similarity searches in this context.

Conclusions:

• Local NB analysis provides insights into the performance of similarity metrics and descriptors at the individual compound level.
• 2D pharmacophore descriptors generally offer better results for similarity-based virtual screening of individual bioactive compounds.
• The choice of descriptor and metric significantly impacts virtual screening success, highlighting the need for careful selection based on local analysis.