A distance function for retrieval of active molecules from complex chemical space representations

Area of Science:

• Chemoinformatics
• Computational chemistry
• Drug discovery

Background:

• Chemical space is crucial for chemoinformatics.
• High-dimensional representations are often considered too complex for compound classification and virtual screening.
• Existing methods may struggle with the complexity of raw chemical spaces.

Purpose of the Study:

• To investigate the effectiveness of a novel distance function in high-dimensional chemical spaces.
• To demonstrate that complex representations are not always necessary for accurate molecular similarity detection.
• To propose a simpler approach for chemoinformatics tasks.

Main Methods:

• Development and application of an "activity-centered" distance function.
• Analysis of molecular similarity relationships in high-dimensional "raw" chemical spaces.
• Testing the function's capability in compound classification and virtual screening contexts.

Main Results:

• The activity-centered distance function accurately detects molecular similarity.
• High dimensionality of chemical space does not impede the function's performance.
• A simple distance metric suffices for complex chemoinformatics tasks.

Conclusions:

• An activity-centered distance function offers a viable and accurate method for analyzing high-dimensional chemical spaces.
• This approach simplifies chemoinformatics research by avoiding complex space representations.
• The findings support the use of simpler metrics for molecular similarity and virtual screening.