Prior Knowledge for Predictive Modeling: The Case of Acute Aquatic Toxicity

Area of Science:

• Environmental toxicology
• Computational chemistry
• cheminformatics

Background:

• Assessing chemical impacts requires toxicological data, often scarce and uncertain.
• Predictive modeling (in silico) estimates properties from existing data.
• Leveraging scientific literature knowledge can enhance predictive modeling.

Purpose of the Study:

• To explore semiautomated knowledge extraction from scientific papers.
• To investigate the use of extracted knowledge in predictive modeling.
• To apply these methods to acute aquatic toxicity.

Main Methods:

• Semiautomated knowledge extraction from scientific literature.
• Application of extracted knowledge to predictive modeling.
• Focus on acute aquatic toxicity data and models.

Main Results:

• Demonstrated utility of knowledge collection and classification.
• Identified benefits for hybrid modeling approaches.
• Showcased improvements in model and predictor selection.

Conclusions:

• Semiautomated knowledge extraction is valuable for predictive toxicology.
• This method aids in addressing data gaps and evaluating model performance.
• Enhances the reliability of chemical safety assessments.