Automated deep lineage tree analysis using a Bayesian single cell tracking approach

Abstract

Single-cell methods are beginning to reveal the intrinsic heterogeneity in cell populations, which arises from the interplay or deterministic and stochastic processes. For example, the molecular mechanisms of cell cycle control are well characterised, yet the observed distribution of cell cycle durations in a population of cells is heterogenous. This variability may be governed either by stochastic processes, inherited in a deterministic fashion, or some combination of both. Previous studies have shown poor correlations within lineages when observing direct ancestral relationships but remain correlated with immediate relatives. However, assessing longer-range dependencies amid noisy data requires significantly more observations, and demands the development of automated procedures for lineage tree reconstruction. Here, we developed an open-source Python library, btrack, to facilitate retrieval of deep lineage information from live-cell imaging data. We acquired 3,500 hours of time-lapse microscopy data of epithelial cells in culture and used our software to extract 22,519 fully annotated single-cell trajectories. Benchmarking tests, including lineage tree reconstruction assessments, demonstrate that our approach yields high-fidelity results and achieves state-of-the-art performance without the requirement for manual curation of the tracker output data. To demonstrate the robustness of our supervision-free cell tracking pipeline, we retrieve cell cycle durations and their extended inter- and intra-generational family relationships, for up to eight generations, and up to fourth cousin relationships. The extracted lineage tree dataset represents approximately two orders of magnitude more data, and longer-range dependencies, than in previous studies of cell cycle heritability. Our results extend the range of observed correlations and suggest that strong heritable cell cycling is present. We envisage that our approach could be extended with additional live-cell reporters to provide a detailed quantitative characterisation of biochemical and mechanical origins to cycling heterogeneity in cell populations.

Learning the Rules of Cell Competition without Prior Scientific Knowledge

“…To do this, we segmented the time-lapse image data using a fully convolutional residual U-Net [20], then used a dedicated convolutional neural network (CNN) to classify each nucleus into one of five states (interphase, prophase, metaphase, anaphase or apoptotic) based on image features [8]. Then, we tracked all cells over time [21]. Next, we classified the fate of each track as either mitotic, apoptotic or unknown using a dedicated cell fate classification network ( Extended Fig 1 , Supplementary Information ).…”

Methodology for comprehensive cell-level analysis of wound healing experiments using deep learning in MATLAB

“…Often, the focus is on tracking cells, generating cell paths and deriving information about cell interactions. For example, Ulicna et al developed an analysis tool (DeepTree) consisting of two neural networks, one for segmentation (U-net) and the other for deciding the status of the cell, like mitosis and apoptosis [ 19 ]. Other groups focus more on segmentation accuracy by developing and establishing new network structures [ 20 – 23 ].…”

Moving beyond velocity: Opportunities and challenges to quantify immune cell behavior*

“…These values are dependent on how fast and far T cells will migrate in the specific imaging volume. An alternative approach, bTrack, 81 uses a Kalman filter to learn and predict future object positions, and a Bayesian believe matrix to assign probabilities that objects belong to a track or will start a new track. The authors of this method utilized this python framework to track cell cycle progression in vitro of the mammalian cell line Madin‐Darby Canine Kidney cells (MDCK) 81 .…”