The Power of Machine Learning Unravels Immune System Interactions
Therapeutic antibodies play a crucial role in treating diseases by modifying immune cells. Scientists from the Institute of AI for Health at Helmholtz Munich in collaboration with Roche present an open-source framework for comprehensive antibody engineering and prediction using imaging flow cytometry data, along with the largest publicly available dataset of human immunological synapse images. This enables in-depth analysis of morphological changes and predictive modeling of T cell cytokine production, offering significant potential in impacting antibody design and streamlining analysis workflows. The new findings are published in Nature Communications.
Therapeutic antibodies change how the T cells in our immune system interact with other cells, making them crucial for various medical treatments. However, it is often unclear how exactly these antibodies influence cell-to-cell interactions.
Researchers around Dr. Carsten Marr from Helmholtz Munich in a close partnership with Roche have tackled this question by capturing and analyzing over 3 million images of interacting cells. Using explainable artificial intelligence (AI), the scientists can measure the changes in these interactions caused by specific antibodies. With the new framework scifAI (single-cell imaging flow cytometry AI) therapeutic antibody functionalities can be predicted in vitro. The approach involves analyzing immunological synapse morphologies early in the immune response initiation and linking them to downstream T cell responses. These characteristics identified in the images can be used to predict the activation of T cells and, consequently, the effectiveness of the antibody.
In addition, the researchers are releasing the largest dataset of interacting cells to date, along with their machine learning model. This significantly contributes to the development of accurate predictions regarding the function of antibodies in the future. By combining high-throughput imaging with data processing and machine learning, this approach holds promise in pharmaceutical research, facilitating the screening of new antibody candidates and ultimately speeding up the development of new therapy options.
Shetab Boushehri et. al. (2023): Explainable Machine Learning for Profiling the Immunological Synapse and Functional Characterization of Therapeutic Antibodies. Nature Communications. DOI: http://dx.doi.org/10.1038/s41467-023-43429-2
About the scientists
Ali Boushehri, doctoral researcher at the Munich School for Data Science (MUDS), at the Institute of AI for Health at Helmholtz Munich and at Roche
Showcases the successful collaboration between Roche and Helmholtz Munich via MUDS.