Skip to content

Jaccard Similarity Coefficient

Summary

The Jaccard Similarity Coefficient, also known as the Jaccard index or binary Jaccard score, is a statistic used to measure the similarity and diversity of sample sets. In cheminformatics, it is widely used to evaluate the similarity between a predicted molecular fingerprint and a ground-truth molecular fingerprint for compound identification and structure elucidation.

Formula

For two binary fingerprint vectors $A$ and $B$, the binary Jaccard score is defined as:

$$J(A, B) = \frac{|A \cap B|}{|A \cup B|} = \frac{N_{11}}{N_{11} + N_{10} + N_{01}}$$

Where: - $N_{11}$ is the number of bits where both $A$ and $B$ have a value of 1. - $N_{10}$ is the number of bits where $A$ is 1 and $B$ is 0. - $N_{01}$ is the number of bits where $A$ is 0 and $B$ is 1.

The score ranges from 0 (no overlap) to 1 (complete identity).

Application in Mass Spectrometry ML

  • Evaluation of Fingerprint Prediction: Machine learning tools like MIST and DreaMS are evaluated by calculating the binary Jaccard score between predicted Morgan fingerprints and the true molecular fingerprints of the test compounds.
  • Performance Benchmarking: It serves as the primary metric for comparing different splits (random vs. scaffold splits) in benchmarks like MassSpecGym to assess the generalization capability of models.

Citations