Exposure assessment modeling

Understanding the fate and transport of contaminants as well as related mechanisms in the environment is essential to  evaluate potential risks of environmental exposures on human health. I develop novel machine learning models to accurately predict  important fate and transport properties of environmental contaminants such as bioconcentration factors and dissipation half-lives.

Find more in these publications:
Gao F, Shen Y, Sallach JB, Li H, Liu C*, Li Y*. (2021). Direct prediction of bioaccumulation of organic contaminants in plant roots from soils with machine learning models based on molecular structures. Environmental Science & Technology 55(24):16358-16368.

Computational toxicology

I develop novel deep learning methods such as supervised autoencoder model and unsupervised graph  learning method for toxicity prediction.1-3 The proposed supervised autoencoder model can learn latent space chemical  representation from hundreds of chemical physicochemical properties and improve prediction accuracy.  Besides chemical physicochemical properties, molecular graph is another popular approach of representing molecules.

Graph learning

Graph structured data are ubiquitours in environmental health sciences. I develop novel graph learning methods such as geometric scattering transform  to learn numerical representation of moelcular graphs, social networks, and microbial networks.

Environemntal epidemiology

Evaluating the potential impacts of exposures on human health is another important  component of my research. Metal exposures have been associated with many adverse health outcomes, but their effects on  multi-omics human gut microbiome were rarely studied. Integrating microbiome metagenomics and metatranscriptomics data, I study the relationships between metal exposures and microbiome species as well as active functional pathways