Robust extraction of cell migration parameters from single-cell trajectories

Abstract

Quantitative analysis of single-cell migration is crucial for elucidating physiological and pathological processes from embryonic development to cancer metastasis. Although the persistent random walk (PRW) model serves as a standard framework for stochastic cell motility, reliable extraction of its core parameters, i.e., migration speed ( S ) and persistence time ( P ), from experimental trajectories remains difficult. Substantial variability in reported S and P values stems from, methodological differences, measurement noise, and biological heterogeneity. Here, we introduce PRW-PIPE, a comprehensive computational pipeline for robust PRW parameter extraction. Benchmarked with simulated trajectories, we characterize observation-time-dependent biases in three prevalent conventional approaches including turning angle (TA) thresholding, velocity autocorrelation function (VACF) analysis, and mean square displacement (MSD) fitting, identifying their characteristic failure regimes. We propose a hybrid method enabling accurate recovery of P across wide ranges. To mitigate the effect of experimental noise, we estimate the noise level based on the MSD method, and incorporate a Kalman filter–based denoising module with a composite metric for parameter refinement, recovering near-ground-truth values. Additionally, we establish an empirical scaling relationship between speed distribution broadening and underlying S and P , supporting model-constrained subpopulation classification via Bayesian information criterion and expectation–maximization clustering. Application to breast cancer cell migration data reveals distinct modulation of motility parameters and subpopulation structure by extracellular matrix composition. This framework outperforms the conventional methods, providing a noise-resilient, reproducible tool for quantitative single-cell motility analysis, with broad utility in mechanistic and high-throughput screening studies.