iSIM-sigma: efficient standard deviation calculation for molecular similarity | JoVE Visualize

Area of Science:

Cheminformatics
Computational Chemistry
Bioinformatics

Background:

Molecular similarity metrics are crucial for cheminformatics tasks like chemical space exploration and subset selection.
Calculating the variance of a complete similarity matrix has quadratic complexity (O(N^2)), which is computationally infeasible for large molecular libraries.
Existing methods struggle with the scalability demands of modern, large-scale molecular datasets.

Purpose of the Study:

To develop computationally efficient methods for calculating the standard deviation of molecular similarities.
To address the limitations of quadratic complexity in analyzing large molecular datasets.
To provide accurate approximations for similarity variance applicable to various similarity indexes.

Main Methods:

Developed an alternative method with O(NM^2) complexity for exact standard deviation calculation of Russell-Rao (RR) and Sokal-Michener (SM) similarity indexes.
Proposed a highly accurate approximation with linear complexity (O(N)) based on sampling representative molecules.
Demonstrated the applicability of the approximation method to other similarity indexes, including Jaccard-Tanimoto (JT).

Main Results:

The proposed approximation method achieves an Root Mean Square Error (RMSE) lower than 0.01 for sets up to 50,000 molecules using only 50 sampled molecules.
The linear complexity approximation significantly outperforms random sampling in accuracy for estimating similarity standard deviation.
The O(NM^2) method provides an exact calculation for RR and SM indexes, offering a more feasible alternative to the O(N^2) pairwise approach.

Conclusions:

The developed approximation method offers a scalable and accurate solution for estimating molecular similarity variance in large datasets.
This approach significantly reduces computational burden, enabling efficient chemical space exploration and subset selection.
The method's accuracy and broad applicability make it a valuable tool for modern cheminformatics research.