Project 2:
Large-Scale Frontier Model for High-Dimensional Biological Imaging Data

Supervisor:
Type:
Requirements:

Advances in bioimaging methods enable the spatially resolved molecular mapping of tissue at single-cell resolution. The resulting data offers a promising opportunity to advance our understanding of tissue structure and function in both health and disease. Such understanding is highly relevant as for instance in oncology the composition and spatial organization of the tumor microenvironment have been shown to directly affect cancer progression, treatment response, and survival. However, the complexity and high dimensionality of the data make its analysis and interpretation by humans or traditional quantitative and computational methods challenging.

In this project, we will develop a large-scale foundation model to extract biologically and clinically relevant information from high-dimensional imaging data. This includes developing novel transformer architectures, designing training algorithms that incentivize the model to detect biological features and their (relative) localization while accounting for the highly heterogeneous nature of diseased tissue through a probabilistic model.