ccbmlib - a Python package for modeling Tanimoto similarity value distributions.

Area of Science:

• Computational chemistry
• Cheminformatics
• Statistical modeling

Background:

• Molecular similarity is crucial in drug discovery and chemical informatics.
• Tanimoto coefficients are widely used to quantify molecular similarity.
• Different molecular fingerprints can yield varying similarity results.

Purpose of the Study:

• To introduce the ccbmlib Python package for modeling similarity distributions.
• To enable statistical assessment of Tanimoto coefficients across diverse fingerprint types.
• To evaluate how molecular similarity is represented by different fingerprint methods.

Main Methods:

• Utilizing the ccbmlib Python package to model Tanimoto coefficient distributions.
• Applying p-values for quantitative comparison of similarity scores.
• Modeling conditional similarity distributions for ranked similarity searches.
• Statistical analysis of feature distributions and correlations in fingerprints.

Main Results:

• Developed accurate models for 11 RDKit fingerprints using ChEMBL data.
• Achieved high accuracy with differences of 1% or less in Tanimoto coefficients for high similarity.
• Demonstrated the utility of p-values for comparing diverse fingerprint representations.
• Conditional significance scores effectively estimate compound rankings in similarity searches.

Conclusions:

• The ccbmlib package provides robust statistical tools for analyzing molecular similarity.
• It facilitates quantitative comparisons and ranking estimations across various fingerprint types.
• The models show high fidelity in representing molecular similarity, aiding cheminformatics applications.