Many artificial tissues are currently being used as disease models for pre-clinical drug development. However, their use is limited by how well they can mimic a given condition. Current methods for fabricating artificial tissues are based on image-derived, or manually designed tissue architectures, and lack high-throughput applicability. This project is developing generative methods for designing bio-printable lung tissues across a spectrum of disease severity in the specific context of mouse and human lung disease.
How can machine learning methods help take the most out of artificial tissues as models for disease? Read in this week’s Helmholtz AI project showcase how researchers at the Max Delbrück Center for Molecular Medicine (MDC) and Helmholtz Munich are working on this challenge.
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
My name is Kyle Harrington and I am the Platform Lead for Image Data Analysis at the Max Delbrück Center for Molecular Medicine in the Helmholtz Association (MDC). I am a computer scientist who works in computational biology, image analysis, and machine learning.
Currently, I am co-PI of the project entitled as Generative Lung Architecture Modelling (GLAM). I developed this project along with my co-PI Dr. Gerald Burgstaller from Helmholtz Munich´s Comprehensive Pneumology Center/Institute of Lung Biology and Disease, as well as our co-Investigator Dr. Dagmar Kainmueller from MDC. Additionally, Deborah Schmidt from MDC is now making key contributions to the project.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
GLAM is developing machine learning strategies to create tunable generative models of lung disease, covering specifically the key aspects of emphysema and fibrosis. Our goal is to create accurate 3D printed models that can used as biological surrogates for pre-clinical drug screening.
There are significant challenges in developing these machine learning strategies.
- The acquisition of high-resolution 3D imaging data from both mouse and human tissue is a challenge that the HMGU team has been expertly navigating.
- On the computational/AI side of the project, we are developing a pipeline for image segmentation that can create 3D bioprintable lung structures, which is a strong expertise of the MDC team.
- One of the highest risk components of the project involves developing a tunable generative model that can produce 3D lung structures parameterized by disease severity.
The potential impact of developing this tool is immense, because a successful outcome means that we will have a highly standardized way of generating and 3D printing a spectrum of lung structures that can be used for pre-clinical screening of drugs. By being able to do this for both mouse and human tissue, we hope that this machine learning-based strategy will be able to provide a bridge between in vitro, in vivo, and clinical research.
How important has the Helmholtz AI funding and platform been to carry out this project?
Helmholtz AI funding enabled us to assemble a brilliant team for this endeavor.
- At HMGU we are supporting Jeanine Pestoni, a PhD student who is focused on the imaging and biological aspects of the project.
- At MDC we are supporting Leo Epstein, a PhD student who is focused on developing machine learning methods for 3D biological structures, and Jan Philipp Albrecht, a PhD student and Research Software Engineer who is focused on machine learning methods and tools for image data.
We have been able to support the development of our machine learning methods for image segmentation, as well as generative model development. The funding has also been critical for acquiring 3D imaging data across the spectrum of disease models to use for training, testing, and validating all our machine learning models.
Without Helmholtz AI’s support, this ambitious project would have been unlikely to have been developed.
Figure: A 3D rendering from the image segmentation of diseased lung tissue from an initial version of the GLAM machine learning pipeline. Data was acquired by Jeanine Pestoni (HMGU), image segmentation was performed by Leo Epstein (MDC), and the 3D visualization was created by Deborah Schmidt (MDC).