Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

BP: Fachverband Biologische Physik

BP 4: Computational Biophysics I

BP 4.2: Vortrag

Montag, 18. März 2024, 15:15–15:30, H 0112

Leveraging Point Cloud Transformers and Simulation-based Inference for Enhanced Parameter Inference in Tumor Growth Modeling — •Julian Herold¹, Eric Behle², and Alexander Schug² — ¹Kalrsruhe Institut für Technologie (KIT), Karlsruhe, Germany — ²Jülich Supercomputing Centre (JSC), Jülich, Germany

Computational modeling serves as a cornerstone in unraveling the intricate dynamics of living tissues. However, the challenge of deriving quantitatively meaningful parameters from experimental data persists. Conventional methods, such as ABC, rely on summary statistics, introducing inherent limitations in the selection of relevant metrics. To address these challenges, we advocate for the adoption of Simulation-based Inference (SBI), harnessing the capabilities of deep learning techniques to navigate the complexities associated with parameter inference. In this study, we present utilizing point cloud transformers directly on positional data extracted from in-vitro spheroids, circumventing the reliance on summary statistics and thus overcoming the limitations of traditional methods. Our methodology integrates the training of neural networks into the parameter inference pipeline of CellsInSilico (CiS), a high-performance framework designed for large-scale tissue simulations. Not only does this yield superior results in terms of inference accuracy, but it also enhances computational efficiency compared to conventional methodologies, empowering researchers to explore critical biological questions. Demonstrated utility includes investigating the interplay between the extracellular matrix and tumor invasion.

Keywords: Simulation-based Inference; Tissue Simulations; Machine Learning