Using Machine Learning Methods and Structural Alerts for Prediction of Mitochondrial Toxicity.

Area of Science:

• Biochemistry
• Toxicology
• Computational Chemistry

Background:

• Organ and idiosyncratic toxicities are frequently associated with mitochondrial toxicity.
• Existing assays for mitochondrial toxicity are well-established but time-consuming for large-scale screening.
• In silico methods offer a promising approach for early hazard identification in drug development pipelines.

Purpose of the Study:

• To develop and validate a computational model for predicting mitochondrial toxicity.
• To create the largest publicly available dataset on mitochondrial toxicity by integrating multiple endpoints.
• To assess the utility of physicochemical properties and structural alerts in predicting mitochondrial toxicity.

Main Methods:

• Compilation of a comprehensive dataset on mitochondrial toxicity by combining multiple experimental endpoints.
• Analysis of physicochemical properties of compounds to identify distinguishing features of toxic molecules.
• Application of machine learning algorithms and structural alerts for in silico risk assessment.

Main Results:

• Physicochemical properties were found to effectively distinguish molecules associated with mitochondrial toxicity.
• The combined approach of machine learning and structural alerts demonstrated high suitability for risk assessment.
• The derived dataset represents the largest collection of mitochondrial toxicity data to date.

Conclusions:

• In silico prediction of mitochondrial toxicity is feasible and beneficial for early-stage drug development.
• Physicochemical properties and structural alerts are key indicators for identifying mitochondrial toxicants.
• This approach facilitates efficient screening of large compound libraries, reducing development risks.