Adaptive AI-Assisted Single-Particle Tracking in Living Cells

Abstract

Single-particle tracking (SPT) offers excellent spatiotemporal resolution for revealing biomolecular structures and functions, but linking complex, multidimensional trajectories to underlying biophysical mechanisms remains challenging in living cells. We introduce the Adaptive AI-assisted Single-Particle Tracking (AAISPT) framework, an automated solution for rapid, accurate processing of multidimensional SPT data sets. AAISPT integrates an adaptive segmentation network (Adap_Seg) for detecting biophysically meaningful trajectory transitions and a pretrained classification model (Adap_Cls). Adap_Cls maps multidimensional features to motion states using diffusion fingerprints and a Transformer encoder, and it was generalized with minimal fine-tuning. AAISPT was benchmarked on synthetic data sets against representative methods from the first AnDi challenge, demonstrating competitive performance in trajectory segmentation and classification. Cross-platform validation shows that AAISPT can reliably identify diverse motion states and resolve dynamic transitions across single-particle trajectories of varying dimensionality. Finally, AAISPT was utilized to analyze ligand-dependent nanoparticle–membrane interactions, validating its capabilities in elucidating complex dynamic processes in living cells.