Segment-based task-specific deep learning framework for detecting state transitions in anomalous diffusion trajectories

Abstract Diffusion in biological systems is inherently complex due to heterogeneous interactions, which lead to anomalous diffusion where single-particle trajectories transition between distinct diffusive states. Detecting these transitions is crucial for understanding the underlying mechanisms, as they offer valuable insights into changes in microscopic interactions. Recent advanced artificial neural networks, known for their remarkable performance across multiple scientific disciplines, have been applied to these detection tasks. In this study, we propose a deep learning framework that leverages change-point detection models and task-specific models to iteratively segment trajectories into current states and residuals and then subsequently characterize the diffusion properties, respectively. The results demonstrate the potential of deep learning approaches for effectively analyzing interactions in complex systems.