Enhancing Acute Oral Toxicity Predictions by using Consensus Modeling and Algebraic Form-Based 0D-to-2D Molecular...

Area of Science:

• Computational Chemistry
• Toxicology
• Drug Discovery

Background:

• Accurate prediction of acute oral toxicity (AOT) is crucial for chemical safety and drug development.
• Existing Quantitative Structure-Activity Relationship (QSAR) models require continuous improvement for reliability and broader applicability.
• The QuBiLS-MAS framework offers a novel approach for molecular encoding in QSAR modeling.

Purpose of the Study:

• To develop and validate robust QSAR models for predicting AOT using the QuBiLS-MAS framework.
• To compare the performance of developed consensus models against existing literature methods.
• To assess the utility of these QSAR models for chemical safety labeling and virtual screening.

Main Methods:

• Employed the QuBiLS-MAS framework for molecular representation.
• Utilized k-nearest neighbor, multilayer perceptron, random forest, and support vector machine algorithms to build base models.
• Developed consensus models (M19, M22, M24) using minimum and weighted average operators, validated on EPA, ProTox, and T3DB datasets.

Main Results:

• Developed three consensus QSAR models (M19, M22, M24) with high predictive accuracy.
• Model M22, built on the EPA-full training set, exhibited the best overall performance across external validation sets (ProTox, T3DB).
• Model M22 showed superior ability in classifying toxic compounds according to the Globally Harmonized System (GHS).

Conclusions:

• The developed QSAR models, particularly M22, are effective tools for predicting AOT.
• These models show significant promise for prospective use in chemical safety assessment and regulatory labeling.
• A freely available software tool (http://tomocomd.com/apps/ptoxra) has been developed for virtual screening applications.