Combinatorial preferences affect molecular similarity/diversity calculations using binary fingerprints and Tanimoto...

Area of Science:

• Computational Chemistry
• Cheminformatics
• Data Analysis

Background:

• Binary fingerprints (FP) are crucial for representing chemical structures in computational studies.
• The Tanimoto coefficient (Tc) is a widely used metric for quantifying similarity between FPs.
• Understanding the statistical distribution of Tc is essential for accurate similarity assessments.

Purpose of the Study:

• To develop a combinatorial method for calculating complete Tanimoto coefficient (Tc) distributions for binary fingerprints (FP).
• To investigate the statistical properties of Tc distributions irrespective of the chemical information encoded in the FPs.
• To analyze the impact of these statistical preferences on average Tc values in large compound databases.

Main Methods:

• Development of a combinatorial approach to compute theoretical Tc distributions.
• Calculation of Tc distributions for binary FPs up to 67 bit positions.
• Application of the method to analyze Tc distributions within a large chemical compound database using various FPs.

Main Results:

• The study successfully calculated complete Tc distributions for binary FPs.
• Theoretical analysis revealed significant statistical preferences for certain Tc values.
• Empirical analysis on a large compound database confirmed these statistical effects, showing they are independent of FP bit length or chemical content.
• Average Tc values are demonstrably biased by these statistically preferred values.

Conclusions:

• A novel combinatorial method provides a comprehensive understanding of Tc distributions.
• Statistical preferences in Tc values inherently bias similarity calculations.
• These findings necessitate a critical re-evaluation of average Tc as a reliable similarity measure without considering underlying distributions.