Text Box:

                                      TASIC: Temporal Assignment of SIngle Cells 

TASIC uses a Hidden Markov Model (HMM) based on a probabilistic Kalman Filter approach to combine time and expression information for determining the branching process associated with time series single cell studies. Once a branching model is determined, each cell is associated with a state in the model and so the different expression trajectories for each cell fate can be reconstructed. As we show by applying our model to myoblast differentiation and

lung development data, using the reconstructed cell fate trajectories we can identify key genes involved in the differentiation process for the different fates. In addition, the learned models can be used to infer functional assignment of cells and derive insights about the synchronization of the process being studied.
























Fig. Penalized likelihood scores for different branching models. a) Structure of 9 of the models tested using our method, each containing at least 3 states. The 9 models include the highest scoring models for the two datasets and a number of other high scoring models out of the 25 models we tested (all models with at most 7 states), see Supplement for complete list of models and their penalized likelihood scores. b) BIC scores obtained for these 9 models for the two datasets we studied in this paper. As can be seen, the optimal models greatly differed between the two datasets indicating the ability of the method to determine an accurate model for a specific set of trajectories.