Prediction of Dynamic Toxicity of Nanoparticles Using Machine Learning

Area of Science:

• Nanotechnology
• Biomedical Engineering
• Computational Toxicology

Background:

• Nanoparticle toxicity prediction is crucial for developing safe biomedical technologies and drugs.
• Assessing nanoparticle safety identifies potential risks to organisms and the environment.
• Machine learning (ML) offers advanced methods for predicting nanoparticle toxicity in various media.

Purpose of the Study:

• To conduct a comparative analysis of ML methods for nanoparticle toxicity assessment.
• To develop and evaluate ML models for predicting nanoparticle toxicity in nutrient solutions.
• To determine optimal nanoparticle dosages and identify key predictive features.

Main Methods:

• Trained a regression model to predict quantitative nanoparticle toxicity based on concentration.
• Developed a multi-class classification model to predict nanoparticle toxicity categories.
• Utilized metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Accuracy, Recall, F1-Score, and Log Loss for model evaluation.

Main Results:

• Regression model achieved MSE = 2.19 and RMSE = 1.48.
• Classification model demonstrated high performance with Accuracy = 0.9756, Recall = 0.9623, F1-Score = 0.9640, and Log Loss = 0.1855.
• Identified optimal dosages: ZnO = 9.5 × 10^-5 mg/mL, Fe₃O₄ = 0.1 mg/mL, SiO₂ = 1 mg/mL.

Conclusions:

• The trained ML models exhibit strong predictive capabilities for nanoparticle toxicity.
• Nanoparticle diameter and concentration in nutrient solutions are the most significant predictive features.
• Established optimal dosage ranges contribute to safer application of nanoparticles in biomedical fields.