This project aims to make machine learning models for biology more interpretable and useful for discovery. We will develop new deep learning methods that combine different types of biological data – such as measurements of genes, proteins, metabolites, and the microbiome – with existing biological knowledge about how these molecules interact in the body. Unlike traditional “black-box” models, our approach will build this biological knowledge directly into the model, so that its inner workings reflect real biological processes. The models will also be able to learn new connections when current knowledge is incomplete.

We will apply these models to study liver disease to uncover key molecular patterns linked to disease severity and outcomes. By testing the models across different patient cohorts and large population datasets, we will ensure that they are robust and broadly applicable. Ultimately, this work will help researchers better understand complex diseases and support the development of more precise diagnostics and treatments.