lazar: a modular predictive toxicology framework.

Area of Science:

• Computational toxicology
• cheminformatics
• predictive modeling

Background:

• Predictive toxicology is crucial for assessing chemical safety.
• Existing methods like read-across have limitations.
• A modular framework can enhance flexibility and accuracy.

Purpose of the Study:

• To introduce LAZAR (Lazy Structure-Activity Relationships), a novel modular framework for predictive toxicology.
• To describe the architecture and capabilities of the LAZAR framework.
• To evaluate the performance of classification and regression models developed using LAZAR.

Main Methods:

• LAZAR employs a modular design, allowing users to select various algorithms for descriptor calculation, similarity assessment, and model building.
• It generates local Quantitative Structure-Activity Relationship (QSAR) models tailored to specific compounds, akin to the read-across approach.
• The framework supports diverse machine learning algorithms for model development.

Main Results:

• The LAZAR framework demonstrates flexibility in constructing predictive toxicology models.
• Example classification and regression models built with LAZAR show promising performance.
• The modularity allows for optimization of QSAR model development for specific toxicological endpoints.

Conclusions:

• LAZAR provides a versatile and customizable platform for predictive toxicology.
• The framework facilitates the development of accurate local QSAR models.
• LAZAR has the potential to improve toxicological risk assessment by offering adaptable modeling strategies.