Dresden 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
BP: Fachverband Biologische Physik
BP 14: Poster Session II
BP 14.46: Poster
Tuesday, March 10, 2026, 18:00–21:00, P2
Physics Informed Neural Networks for Microbial Interaction Network Inference — •Luca Battiston1,2 and Frank Cichos1 — 1Universität Leipzig, Linnestr. 5, 04103, Leipzig, Germany — 2Helmholtz Centre for Environmental Research GmbH (UFZ), Permoserst. 15, 04318, Leipzig, Germany
Recent advances in machine learning have enabled data-driven modeling of complex dynamical systems, with growing interest in methods that extract meaningful information about the underlying physical laws. Among these, Physics-Informed Neural Networks (PINNs) have emerged as a powerful tool for system identification, particularly in settings where data are scarce or noisy. On the other hand, generalized Lotka-Volterra (gLV) equations are widely used to model microbial community dynamics, provided that the underlying interaction matrix is known. In this work, we aim to reconstruct these interactions by combining PINNs with Least Squares Regression for inference of the interaction network. We evaluate our method using extensive gLV simulations covering a range of interaction matrix complexities and noise levels. Our results demonstrate high accuracy in network recovery and show that the approach retains robustness under measurement noise. This provides a step toward developing a robust and flexible framework for identification of complex interaction patterns in microbial communities.
Keywords: Machine Learning; Interaction Networks; Physics Informed Neural Networks; Microbial Communities; System Identification