Machine Learning Toxicity Prediction: Latest Advances by Toxicity End Point.

Area of Science:

• Computational toxicology
• cheminformatics
• drug discovery

Background:

• Machine learning (ML) models are increasingly used for predicting small molecule toxicity.
• Computational toxicity prediction aids early-stage drug discovery by filtering potential drug candidates.
• Advancements in toxicology databases support ML applications, yet understanding ML scope and applicability remains crucial.

Purpose of the Study:

• To review recent progress and challenges in ML-based toxicity prediction.
• To provide a detailed description of state-of-the-art ML models for various toxic endpoints.
• To analyze molecular representation, algorithms, and evaluation metrics used in ML toxicity studies.

Main Methods:

• Literature review focusing on recent advancements in ML for toxicity prediction.
• Analysis of ML models applied to specific toxic endpoints (e.g., acute oral toxicity, hepatotoxicity, mutagenicity).
• Examination of molecular representations, algorithms, and evaluation metrics in published research.

Main Results:

• Detailed overview of ML model performance across different toxic endpoints.
• Identification of common molecular representations, algorithms, and evaluation metrics.
• Highlighting of challenges in comparing ML algorithm performance due to data complexity and chemical space variations.

Conclusions:

• ML models show promise for predicting diverse toxic endpoints, but performance varies.
• Further research is needed to understand the scope and limitations of ML methods in toxicology.
• Standardized evaluation and reporting are essential for robust ML-driven toxicity assessments.