Rough sets and social ski-driver optimization for drug toxicity analysis.

Area of Science:

• * Computational toxicology
• * Zebrafish embryo models
• * Machine learning in drug development

Background:

• * Manual toxicity testing using zebrafish embryos is time-consuming and not feasible for large-scale studies.
• * Developing automated, accurate methods for drug toxicity assessment is crucial for new drug development.
• * This research addresses the need for an efficient automated model for evaluating toxicant effects.

Purpose of the Study:

• * To develop and validate an automated model for investigating the toxicity of chemical compounds using zebrafish embryos.
• * To employ advanced feature extraction and selection techniques for accurate toxicity classification.
• * To compare the performance of the proposed model against existing algorithms.

Main Methods:

• * Feature extraction from zebrafish embryo images using Segmentation-Based Fractal Texture Analysis (SFTA).
• * Application of a novel rough set-based model with Social Ski Driver (SSD) for minimal feature subset selection.
• * Classification of embryo viability (alive or coagulant) using the AdaBoost classifier.

Main Results:

• * The proposed automated model achieved high classification performance, ranging from 97.1% to 99.5%.
• * The model demonstrated superior performance compared to three deterministic rough set reduction algorithms and a PSO-based algorithm.
• * Successful identification of toxic effects of 3, 4-Dichloroaniline (34DCA) and p-Tert-Butylphenol (PTBP).

Conclusions:

• * The developed drug toxicity model is highly efficient, particularly in rough set-based feature selection.
• * The model achieves excellent classification performance, offering a reliable tool for toxicity assessment.
• * This automated approach provides a feasible alternative to manual toxicity testing in drug development.