Development of a QSAR model to predict hepatic steatosis using freely available machine learning tools

Area of Science:

• Toxicology
• Computational Chemistry
• Hepatology

Background:

• Hepatic steatosis, particularly non-alcoholic fatty liver disease (NAFLD), is a prevalent health concern often linked to chemical and environmental exposures.
• Existing predictive models focus on molecular initiating events, but lack prediction for the apical effect of steatosis.
• Quantitative structure-activity relationship (QSAR) modeling offers a rapid and cost-effective approach for identifying steatogenic compounds.

Purpose of the Study:

• To develop and validate a QSAR model for predicting hepatic steatosis.
• To identify chemical descriptors associated with steatosis induction.
• To provide a tool for screening pharmaceuticals and pesticides for steatogenic potential.

Main Methods:

• Development of a QSAR model using freely available machine learning tools.
• Utilized a dataset of 207 pharmaceuticals and pesticides classified as steatotic or non-steatotic based on in vivo studies.
• Employed linear discriminant analysis (LDA) within the TANAGRA software.

Main Results:

• The best performing LDA model achieved 70% accuracy, 66% sensitivity, and 74% specificity.
• The model was based on four significant chemical descriptors.
• The model demonstrates potential for predicting steatosis from chemical structure.

Conclusions:

• A QSAR model was successfully developed to predict hepatic steatosis with reasonable accuracy.
• This model can serve as a valuable tool for early identification of steatogenic compounds.
• Expansion of the dataset to include diverse chemical types will enhance future model development and broaden applicability.