ToxiM: A Toxicity Prediction Tool for Small Molecules Developed Using Machine Learning and Chemoinformatics...

Area of Science:

• Computational chemistry
• Toxicology
• Drug discovery

Background:

• Experimental methods for predicting molecular toxicity are laborious and time-consuming.
• Computational approaches can provide efficient alternatives for toxicity assessment.

Purpose of the Study:

• To develop a computational tool for predicting molecular toxicity, aqueous solubility, and permeability.
• To leverage machine learning for accurate prediction of these crucial molecular properties.

Main Methods:

• Utilized a curated dataset of toxin molecules for training machine learning models.
• Exploited chemical and structural features (descriptors and fingerprints) for model development.
• Developed both classification models for toxicity and regression models for solubility and permeability.

Main Results:

• Classification models achieved high accuracy (93%) and correlation coefficients (0.84) on cross-validation and blind datasets.
• Random forest regression model showed excellent performance for solubility prediction (R² = 0.84).
• Partial least squares regression model demonstrated good performance for permeability prediction (R² = 0.68).

Conclusions:

• The developed ToxiM web server is a reliable and accurate tool for predicting molecular toxicity, solubility, and permeability.
• Computational predictions can significantly expedite the assessment of small molecule properties in research and development.