RESEARCH
Applying Machine Learning and Artificial Intelligence for Predicting ADME-tox Properties of Environmental Chemicals
Highlights​
-
New PFOS Model: Developed a detailed PBPK model for PFOS during pregnancy and lactation.
-
Improved Risk Estimates: The model predicted lower human equivalent doses (0.08–0.91 μg/kg/day) than current EPA estimates, suggesting existing risk assessments may underestimate fetal and neonatal exposure.
-
Regulatory Tool: Supports refined PFOS risk assessments for sensitive populations.
Research descriptions:
ADME (Absorption, Distribution, Metabolism, and Excretion) and toxicity (tox) are critical factors in assessing the impact of chemicals on living organisms and the environment. This project aims to leverage advanced computational techniques, specifically the physiologically based pharmacokinetic (PBPK) and quantitative structure-activity relationship (QSAR) model, to predict the ADME-tox properties of environmental chemicals and enhance our understanding and prediction capabilities regarding how environmental chemicals interact with biological systems, facilitating more informed and efficient risk assessment in environmental and health contexts. Additionally, the machine learning and artificial intelligence algorithms were further used to support and assist these computational models in implication in PFAS risk assessment and further developed the AI-assisted PBPK model to estimate the tumor delivery efficiency of cancer nanomedicines.
Picture​
​​
Computational Modeling on the Interaction of Nanomedicine with Protein Corona and Its Impact on Tumor Delivery Efficiency
Highlights​
-
AI-Enhanced PBPK Modeling: A new 3-step framework uses AI/ML to predict ADME parameters and integrate them into PBPK models for faster, animal-free pharmacokinetic simulations.
-
Neural-ODE Breakthrough: Neural-ODE models outperform traditional methods in predicting time-series drug concentration profiles, especially under complex dosing.
-
Key Challenges: Interpretability and limited data diversity remain issues, but deep learning offers promising solutions for robust, scalable PK modeling.
Research descriptions:
Motivated by the advanced development of artificial intelligence (AI) models, AI-based approaches have found promising applications across diverse domains, ranging from generative language and image generation to drug discovery. Specifically, generative adversarial networks (GAN) in machine learning have successfully been applied in smart drug design and generate novel compounds with the desired functionality, while neural ordinary differential equations (NODEs) efficiently resolve pharmacokinetic profiles through data-driven learning. This project aims to develop an AI-based approach that integrates GAN and NODE methodologies to optimize interactions between nanoparticles (NPs) and protein corona for enhancing tumor delivery efficiency in nanomedicine. Our hypothesis posits that the combined use of GAN and NODEs will enable the generation of virtual protein corona fingerprints corresponding to distinct physicochemical properties of nanoparticles. Neural ODE, considering the interplay between protein corona and NPs properties, will simulate NPs biodistribution, facilitating the identification of ideal nanomedicine candidates with optimal tumor delivery efficiency.​​
Picture​
​​
Development of Generative AI in Reconstructing the Structure of Unidentified Chemicals and Predicting their Combined Toxicity in Environmental metrics from Non-Target Analysis
Research descriptions:
By employing generative AI, the study aims to enhance the accuracy and efficiency of reconstruction of the molecular structures of unknown substances in environmental samples based on the detected signals from non-target analysis. Furthermore, the project will be extended to predicting the combined toxicity of these unidentified chemicals in environmental metrics. This innovative approach promises to improve our understanding of the environmental impact of unknown chemicals, offering valuable computational tools for effective risk assessment and management.