XML-CIMT: Explainable Machine Learning (XML) Model for Predicting Chemical-Induced Mitochondrial Toxicity.

Area of Science:

• Toxicology
• Computational Chemistry
• Machine Learning

Background:

• Chemicals, including pharmaceuticals, can cause organ toxicity, with mitochondrial dysfunction being a key factor.
• Drug recalls, such as for troglitazone, highlight the critical need to identify and mitigate mitochondrial toxicity.
• Existing methods for predicting chemical-induced mitochondrial toxicity require improvement.

Purpose of the Study:

• To develop and validate an explainable machine learning model for predicting chemical-induced mitochondrial toxicity.
• To identify key molecular features associated with mitochondrial toxicity.
• To provide a tool that aids pharmaceutical chemists in assessing compound safety.

Main Methods:

• Utilized Mordred descriptors for feature extraction after rigorous feature selection.
• Employed the CatBoost machine learning algorithm for classification.
• Validated the model using 10-fold cross-validation and independent testing.

Main Results:

• The selected Mordred features combined with the CatBoost algorithm achieved 85% accuracy in 10-fold cross-validation.
• The model demonstrated 87.1% prediction accuracy on an independent test set.
• The proposed model outperformed existing state-of-the-art methods in predicting mitochondrial toxicity.

Conclusions:

• The developed explainable machine learning model reliably predicts chemical-induced mitochondrial toxicity.
• The model's accuracy surpasses current benchmarks, offering a significant advancement in safety assessment.
• The explainable nature of the model enhances understanding of toxicity mechanisms for pharmaceutical chemists.