Mathias Engel - Research

Across many projects our lab utilizes robotic high-content screening to image millions of cells. I analyze the images and extract cell specific feature vectors. These feature vectors are created to describe the fundamental morphological properties of the cells and are used in further modelling of the cell signalling and behaviour. The algorithmic tools to segment, define and describe cells are still very much a field of development that I want to contribute to.

In order to simultaneously correlate and predict relations between morphologies and various prior information known from the experimental design (cell line, treatment, RNAi) new statistical models are needed. Such models should be capable of combining uncertainties across different data types. In our lab we therefore exploit the properties of deep learning architectures to model the probability distribution of the morphological space across multiple conditions. Deep learning is a machine learning method that consist of hierarchical networks that are trained in a layer by layer fashion. By using layers the method tries to model higher-level abstractions of the input given. With this class of models we aim at discovering the detailed language of cell morphology and correlate this with resistance, invasion and migration.