ISSN print edition: 0366-6352 ISSN electronic edition: 1336-9075 Registr. No.: MK SR 9/7 Published monthly

A graph theoretic and machine learning framework based on Ricker wavelet neural network for QSPR and multi-criteria decision analysis

Wakeel Ahmed, Zarwish Naeem, and Shahid Zaman

Department of Mathematics, University of Sialkot, Sialkot, Pakistan

E-mail: wakeelahmed784@gmail.com

Received: 10 February 2026  Accepted: 9 March 2026

Abstract:

Machine learning techniques provide a powerful data-driven framework for the quantitative evaluation of drug candidates and the enhancement of treatment-oriented decision-making. In this study, a comprehensive computational framework is proposed in which reverse eccentricity and structure dependent topological descriptors of selected drugs are employed as primary descriptors for property prediction by converting chemical structure into a molecular graph. These indices were first computed to encode the intrinsic structural and connectivity information of the compounds and subsequently integrated into Quantitative Structure Property Relationship (QSPR) modeling. Furthermore, machine learning algorithms namely, Ricker Wavelet Neural Network (RWNN) and Random Forest (RF) were employed to estimate key physicochemical properties of the drugs. The standard statistical measure is used to evaluate model performance such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and coefficient of determination (\(R^2\)). The findings show that RWNN performed better than RF especially in the process of identifying more complex nonlinear interactions between molecular topology and physicochemical behavior. In order to facilitate rational prioritization, the predicted properties have been further prioritized by the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), which is a multi-criteria decision-making (MCDM) method ranking candidate compounds by a number of criteria, often competing, and typically involving a weighting factor. The suggested combination of topological indices, machine learning driven QSPR modeling and MCDM is providing a clear and systematic approach to early stage drug evaluation and prioritization, which is potentially useful in the computational drug discovery pipeline.

Keywords: Stomach cancer; Machine learning; Ricker wavelet neural network; QSPR analysis; MCDM

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-026-04794-6

Chemical Papers 80 (6) 6699–6725 (2026)