Extended continuous similarity indices: theory and application for QSAR descriptor selection

Area of Science:

• Computational chemistry
• Cheminformatics
• Data analysis

Background:

• Traditional similarity indices are limited to pairwise comparisons of binary or categorical data.
• N-ary similarity indices offer computational efficiency by comparing multiple objects simultaneously.
• Existing n-ary indices are not suitable for continuous numerical data.

Purpose of the Study:

• To generalize n-ary similarity indices for application to numerical data with continuous components.
• To explore and present the analytical properties of these extended continuous similarity indices.
• To demonstrate the utility of the extended formalism in practical cheminformatics applications.

Main Methods:

• Formulation of generalized n-ary similarity indices for continuous data.
• Analytical derivation and discussion of the properties of the extended indices.
• Application of the extended indices to feature selection in quantitative structure-activity relationship (QSAR) modeling.

Main Results:

• Successful extension of n-ary similarity indices to handle continuous numerical vectors.
• Demonstration of improved computational efficiency (O(N) scaling) compared to quadratic methods.
• Validation of the extended indices in QSAR for effective descriptor set discernment.

Conclusions:

• The generalized n-ary similarity indices provide a powerful and efficient tool for analyzing continuous data.
• This extension broadens the applicability of n-ary indices to a wider range of cheminformatics and data analysis tasks.
• The method offers a convenient approach for feature selection and descriptor analysis in QSAR.