Deep Probabilistic Learning Model for Prediction of Ionic Liquids Toxicity.

Area of Science:

• Computational Chemistry
• Toxicology
• Machine Learning

Background:

• Assessing ionic liquid toxicity is crucial but traditional methods are costly and time-consuming.
• Computational models offer a viable alternative for efficient toxicity prediction.

Purpose of the Study:

• To develop a probabilistic deep kernel learning model for predicting ionic liquid toxicity.
• To utilize open-source tools (RDKit, Mol2vec) for structure-based toxicity prediction.
• To provide uncertainty quantification for model predictions.

Main Methods:

• Developed a probabilistic model using deep kernel learning.
• Employed RDKit and Mol2vec for feature generation directly from chemical structures.
• Validated the model on the leukemia rat cell line (IPC-81).

Main Results:

• Achieved a Root Mean Square Error (RMSE) of 0.228 and R-squared (R2) of 0.943.
• Demonstrated high reliability and accuracy in toxicity predictions.
• Integrated uncertainty estimation to account for experimental data variability.

Conclusions:

• The developed model accurately predicts ionic liquid toxicity based on chemical structure.
• The model offers a cost-effective and efficient alternative to experimental toxicity testing.
• A web server facilitates easy access for researchers to predict ionic liquid toxicity.