Predicting cytotoxicity of engineered nanoparticles using regularized regression models: an in silico approach

Area of Science:

• Environmental Science
• Toxicology
• Computational Chemistry

Background:

• Engineered nanoparticles (NPs) are widely used but their physicochemical modifications can induce toxicity.
• Understanding NP toxicity is crucial for safe application and risk assessment.
• Predictive models can aid in evaluating NP safety profiles.

Purpose of the Study:

• To develop and validate regularized regression models for predicting engineered NP cytotoxicity.
• To identify key physicochemical descriptors influencing NP toxicity.
• To assess the generalization capability of predictive models on independent datasets.

Main Methods:

• Compiled a dataset of engineered NP cytotoxicity from 2010-2022.
• Handled missing data using listwise deletion and kNN imputation, creating two datasets.
• Constructed and validated ridge, LASSO, and elastic net regularized regression models.

Main Results:

• Models achieved high F1 scores, ranging from 91.81% to 93.63% across internal and external validations on both datasets.
• The developed models demonstrated strong generalization to unseen data.
• Cell type, material, cell source, cell tissue, synthesis method, and coat/functional group were identified as critical predictors of NP cytotoxicity.

Conclusions:

• Regularized regression models accurately predict engineered NP cytotoxicity.
• The models are robust and generalize well to new data.
• Key descriptors identified can guide the design of safer nanomaterials.