Inferring TF activation order in time series scRNA-Seq studies

Abstract

Methods for the analysis of time series single cell expression data (scRNA-Seq) either do not utilize information about transcription factors (TFs) and their targets or only study these as a post-processing step. Using such information can both, improve the accuracy of the reconstructed model and cell assignments, while at the same time provide information on how and when the process is regulated. We developed the Continuous-State Hidden Markov Models TF (CSHMM-TF) method which integrates probabilistic modeling of scRNA-Seq data with the ability to assign TFs to specific activation points in the model. TFs are assumed to influence the emission probabilities for cells assigned to later time points allowing us to identify not just the TFs controlling each path but also their order of activation. We tested CSHMM-TF on several mouse and human datasets. As we show, the method was able to identify known and novel TFs for all processes, assigned time of activation agrees with both expression information and prior knowledge and combinatorial predictions are supported by known interactions. We also show that CSHMM-TF improves upon prior methods that do not utilize TF-gene interaction. OPEN ACCESSCitation: Lin C, Ding J, Bar-Joseph Z (2020) Inferring TF activation order in time series scRNA-Seq studies. PLoS Comput Biol 16(2): e1007644. on the method. As we show, the method is able to scale to large and noisy datasets and so is appropriate for several studies utilizing time series scRNA-Seq data.Inferring TF activation order in time series scRNA-Seq studies PLOS Computational Biology | https://doi.

A Bayesian inference transcription factor activity model for the analysis of single-cell transcriptomes

“…For example, both BITFAM and CSHMM-TF inferred increased GATA6 activity in AT2 cells and increased SOX4/SOX5 activity in lung development (Supplemental Fig. S20), consistent with what is known about the mechanistic roles of these TFs in the respective cell types (Poncy et al 2015;Flodby et al 2017;Lin et al 2020). However, the CSHMM-TF model also generates developmental trajectories, whereas BITFAM does not.…”

“…BITFAM can be readily combined with multiple clustering or trajectory building approaches. The CSHMM-TF (Lin et al 2020) combines transcription factor activity inference with the generation of developmental trajectories based on a continuous state hidden Markov model. Although the CSHMM-TF approach is ideally suited for temporal or developmental trajectories involving state transitions, BITFAM can infer transcription factor activities for data sets that do not contain temporal trajectories and state transitions, thus complementing CSHMM-TF.…”

“…We next compared the results of BITFAM to the CSHMM-TF model (Lin et al 2020), which combines transcription factor activity inference with the generation of developmental trajectories based on a continuous state hidden Markov model. We applied the CSHMM-TF and BITFAM models in a lung development data set (Treutlein et al 2014) and found that the inferred TFs activities from BITFAM overlapped with the results from the CSHMM-TF approach (Lin et al 2020). For example, both BITFAM and CSHMM-TF inferred increased GATA6 activity in AT2 cells and increased SOX4/SOX5 activity in lung development (Supplemental Fig.…”

Computational biology approaches for mapping transcriptional regulatory networks

“…Computational biology approaches to map TRNs from these data have to adjust to their inherent stochastic variation and sparsity. These approaches are designed to map TRNs at the level of unique cells [62] , [84] , [85] , [86] or to help reconstruct dynamic cell trajectories or putative TF order of action [87] , [88] , [89] . SCENIC (Single-Cell regulatory Network Inference and Clustering) [62] identifies sets of genes that are co-expressed with TFs, using GENIE3 [39] and the faster variant of it, GRNBoost [40] .…”

“…PIDC (Partial Information Decomposition and Context) [86] uses partial information decomposition to identify potential regulatory relationships between genes, and outputs a weighted network from an expression matrix. Finally, SCODE (scRNA-seq performed on differentiating cells by integrating the transformation of linear ODEs and linear regression) [88] , SCNS (Single-Cell Network Synthesis) [87] and CSHMM-TF (Continuous-State Hidden Markov Models TF) [89] interpret scRNA-seq as time-course expression data, where the pseudo-time corresponds to the time information, and are relevant for biological systems undergoing dynamic transcriptional changes. A recent benchmark was performed on six scRNA-seq network inference methods, primarily developed for bulk RNA-seq, based on their ability to infer similar networks when applied to two independent data sets for the same biological condition [90] .…”