A new paradigm of genomics research.
Genome Organization
Genome organization is crucial for determining gene regulation but are difficult to measure. We built a machine learning model, C.Origami that predicts cell-type-specific genome organization patterns. We further developed in silico genetic screen to identify regulatory elements that are important for genome organization.
DNA-binding Proteins
There are over 2,500 DNA-binding proteins in human. They bind to genomic DNA and directly regulate gene expression. We developed Chromnitron to predict their cell-type-specific binding profiles and identify key proteins that determines human immune cells functions.
Spatial Cellular Organization
To learn principles of cellular organization from multiplexed imaging data, we implemented a self-supervised representation learning approach: CANVAS. We found that the learned features can be utilized to distinguish morphological differences between tumor microenvironments and stratify patients.
