Finch: Toxicity Dose Response Curve Prediction of Chemical Compounds and Mixtures

Abstract

A holistic approach to chemical mixtures is reshaping risk assessment emphasizing mixture testing over single compounds eliminating animal testing and advancing modeling methods. Most computational models still focus on individual chemicals and conventional mixture models like concentration addition and independent action are limited they struggle with multiple Modes of Action and often miss synergistic or antagonistic effects. Regulatory agencies need faster more efficient models that go beyond these constraints. Finch addresses these challenges with a novel workflow. It integrates molecular descriptor based frameworks and deep learning embeddings in multi task quantitative structure activity relationship models for improved chemical exposure prediction. DL embeddings preserve information from diverse inputs molecular descriptors physicochemical properties and large language model embeddings from SMILES by distilling key features into a latent space enhancing machine learning predictions. Finch multi task learning optimizes multiple loss functions simultaneously leveraging all available data to learn generalized representations. This enables effective modeling of complex ingredient interactions within mixtures offering a significant advancement for regulatory safety assessment.