ccbmlib

ccbmlib models Tanimoto coefficient distributions for RDKit molecular fingerprints to enable statistical assessment and significance testing of molecular similarity.

Key Features:

• Tanimoto Coefficient Modeling: Models the distribution of Tanimoto similarity coefficients computed from binary molecular fingerprints.
• Statistical Significance Assessment: Derives significance measures and p-values to compare similarity scores across different fingerprint representations and value ranges.
• Conditional Distribution Modeling: Estimates conditional distributions for a given reference compound to compute conditional significance scores and predict ranking in similarity searches.
• Feature Distribution and Correlation Analysis: Performs statistical analyses of feature distributions and correlations within fingerprint datasets to inform the models.
• RDKit Fingerprint Compatibility: Operates on molecular fingerprints provided by RDKit.
• Evaluation on ChEMBL: Validated on the ChEMBL compound collection using 11 different RDKit fingerprints, reporting modeling differences for high-similarity Tanimoto coefficients within approximately 1%.

Scientific Applications:

• Molecular Similarity Evaluation: Quantitatively evaluates how molecular similarity is reflected across different fingerprint representations.
• Comparative Analysis: Enables quantitative comparison of similarity score distributions derived from various molecular fingerprints.
• Rank Prediction in Similarity Searches: Predicts ranking positions of test compounds in similarity searches via conditional significance scoring, supporting virtual screening analyses.

Methodology:

Uses RDKit to obtain molecular fingerprints; conducts statistical analyses of feature distributions and correlations within fingerprint datasets; develops models of Tanimoto coefficient distributions validated against a ChEMBL reference dataset.