Artificial Intelligence
Quantitative analysis of live-cell imaging data is challenging due to large temporal and spatial datasets as well as diverse experimental settings and analysis tasks. The experimental and computational groups in the initiative collaborate to develop artificial intelligence based methods to meet these analysis needs. This also includes the collection of large training datasets and the application of new methods to support ongoing biomedical research.
Live-Cell Imaging
The group of Constantin Pape is developing methods for the automated analysis of microscopy images using deep learning and computer vision methdos. In particular, his group is focused on methods that require as few training data as possible, since the requirement for (manually created) training examples is one of the key hurdles to applying modern image analysis solutions in practice. In the context of QuCellAI, he is collaborating with several groups to automate quantitative analysis of long-running experiments involving imaging on the Incucyte platforms. The eventual goal is to collect a large and diverse dataset of Incucyte images that will be used to develop versatile algorithms for cell segmentation and cell tracking and can be established as a new benchmark dataset for microscopy image analysis.
Examples for cell segmentation in LiveCell images (left) and tracking in sparse (middle) and crowded (right) situations are shown in the images below. Data kindly provided by Argyris Papantonis and Carmelo Ferrai.
Contact: Prof. Dr. Constantin Pape
Data-driven biophysics of cardiac muscle structure and function
The ability of the cardiac muscle to contract strongly and robustly is crucial for maintaining health. While the molecular basis of muscle contraction is well studied, it is unclear how organized function on organ level emerges across length scales, from intrinsically stochastic protein kinetic over heterogeneous intracellular cytoskeletal dynamics (sarcomeres), to cell-cell interactions in muscle tissue.We use advanced high-throughput microscopy, data science and complexity theory to study the emergence of cardiac contractility in single stem-cell derived cardiac muscle cell and in engineered 3D cardiac muscle tissue. We develop custom AI-based algorithms for the analysis of cardiac muscle structure and dynamics. The obtained detailed data allow us to obtain novel systemic perspectives on cardiac muscle, by (1) inferring complex dynamic models for muscle morphogenesis and motion, (2) better understanding the pathogenesis of cardiomyopathies, and (3) testing the chronic and acute effect of drugs and novel treatment methods.
Movie shows real-time beating dynamics of single human induced pluripotent stem cell-derived cardiac muscle cell (cardiomyocyte) with endogenous fluorescent labeling of sarcomere Z-bands using CRISPR/Cas9 (own data).
Contact: Dr. Daniel Härtter