Machine learning analysis reveals the dynamics of mode transition in dendritic cell migration

Abstract

Dendritic cells (DCs) patrol the body as immunological sentinels and search for pathogens. Upon stimulation, immature DCs (imDCs) become mature DCs (mDCs), which migrate to the lymph nodes and present antigens to T cells. The migratory behavior is crucial for initiating and controlling immune responses; however, the properties of the highly heterogeneous and dynamic motility phenotype are not fully understood. Here, we established an unsupervised machine learning (ML) strategy to investigate spatiotemporal motility patterns in long-term, two-dimensional cell migration trajectories, and determined the number of motility patterns and how these are related to the maturation status. We identified three distinct migratory modes independent of the cell state: slow-diffusive (SD), slow-persistent (SP), and fast-persistent (FP). We found that maturation-dependent motility changes are emergent properties of the distribution and dynamic transitions of these three modes. Remarkably, imDCs changed their migration modes more frequently, and predominantly followed the SD→FP→SP→SD unicyclic transition, indicating that imDCs rapidly increase their speed during the shift from diffusive to persistent motility; however, persistence progressively declines when switching back to diffusive motility. In contrast, mDCs show no transition directionality. Our ML-promoted motility pattern analysis and history-dependent mode transition investigation may provide new insights into the complex process of biological motility.

Ovarian cancer cells exhibit diverse migration strategies on stiff collagenous substrata

“…The most common mathematical metrics used currently for quantifying cellular shape include aspect ratio, eccentricity, shape index, and circularity, while for cell migration, it includes accumulated distance, average velocity, turning angles, and mean square displacement (11,12). Advancement in computational approaches has resulted in several machine learning and neural network-based approaches for cellular segmentation and classification models for distinguishing and characterizing various morphology and migratory modes (13)(14)(15). However, the complexity of such models draws researchers back to using static mathematical parameters to quantify morphology and migration.…”