Analysis of Cell Lineage Trees by Exact Bayesian Inference Identifies Negative Autoregulation of Nanog in Mouse Embryonic Stem Cells

Abstract

Many cellular effectors of pluripotency are dynamically regulated. In principle, regulatory mechanisms can be inferred from single-cell observations of effector activity across time. However, rigorous inference techniques suitable for noisy, incomplete, and heterogeneous data are lacking. Here, we introduce stochastic inference on lineage trees (STILT), an algorithm capable of identifying stochastic models that accurately describe the quantitative behavior of cell fate markers observed using time-lapse microscopy data collected from proliferating cell populations. STILT performs exact Bayesian parameter inference and stochastic model selection using a particle-filter-based algorithm. We use STILT to investigate the autoregulation of Nanog, a heterogeneously expressed core pluripotency factor, in mouse embryonic stem cells. STILT rejects the possibility of positive Nanog autoregulation with high confidence; instead, model predictions indicate weak negative feedback. We use STILT for rational experimental design and validate model predictions using novel experimental data. STILT is available for download as an open source framework from http://www.imsb.ethz.ch/research/claassen/Software/stilt---stochastic-inference-on-lineage-trees.html.

Modeling signaling‐dependent pluripotency with Boolean logic to predict cell fate transitions

“…The concept of dynamic heterogeneity that is explored here is experimentally supported by single-cell tracking studies (Singer et al, 2014;Filipczyk et al, 2015) and has been applied in stochastic ODE models of transcriptional noise (Kalmar et al, 2009), as well as in the inference of stem cell fate (Feigelman et al, 2016). We assume that heterogeneous populations of stem cells correspond to SCCs in the state transition graphs of asynchronously updated Boolean models.…”

Time-dependent propagators for stochastic models of gene expression: an analytical method

“…(2016) suggests that, in the context of Nanog expression in mouse embryonic stem cells, autoregulation rates are indeed small compared to other model parameters.…”

“…(2016) for recent work combining stochastic simulation with a particle filtering approach. However, these approaches can still be very time-consuming, due to the (relatively) high dimensionality of the model parameter space, combined with the fact that, for each combination of parameter values, the stochastic model has to be simulated sufficiently many times to yield a probability distribution that can be used to infer the corresponding propagator.…”

Parameter inference for stochastic single-cell dynamics from lineage tree data

“…Finally, Feigelman et al [17] proposed a method for exact Bayesian parameter inference from cell lineage data that uses particle filtering to approximate the full joint state and parameter posterior distribution. The method was successfully applied to a stochastic gene expression system that is critical for stem cell differentiation and clearly demonstrated the strengths of lineage-based inference.…”

“…To achieve scalability of our method with the number of generations, we make use of a plausible simplifying assumption in the likelihood decomposition which is shown to work well in practice. In contrast to [17], our method allows efficient likelihood calculation and smaller particle degeneracy with increasing tree lengths, which allows us to extract information out of longer lineages. Furthermore, parameter sampling and likelihood approximation are carried out separately from each other, which permits the use of more powerful samplers (such as Population Monte Carlo [18] or Nested Sampling [19]) for the efficient exploration of high-dimensional parameter spaces.…”