Causal network inference from gene transcriptional time-series response to glucocorticoids

Abstract

Gene regulatory network inference is essential to uncover complex relationships among gene pathways and inform downstream experiments, ultimately enabling regulatory network re-engineering. Network inference from transcriptional time-series data requires accurate, interpretable, and efficient determination of causal relationships among thousands of genes. Here, we develop Bootstrap Elastic net regression from Time Series (BETS), a statistical framework based on Granger causality for the recovery of a directed gene network from transcriptional time-series data. BETS uses elastic net regression and stability selection from bootstrapped samples to infer causal relationships among genes. BETS is highly parallelized, enabling efficient analysis of large transcriptional data sets. We show competitive accuracy on a community benchmark, the DREAM4 100-gene network inference challenge, where BETS is one of the fastest among methods of similar performance and additionally infers whether causal effects are activating or inhibitory. We apply BETS to transcriptional time-series data of differentially-expressed genes from A549 cells exposed to glucocorticoids over a period of 12 hours. We identify a network of 2768 genes and 31, 945 directed edges (FDR ≤ 0.2). We validate inferred causal network edges using two external data sources: Overexpression experiments on the same glucocorticoid system, and genetic variants associated with inferred edges in primary lung tissue in the Genotype-Tissue Expression (GTEx) v6 project. BETS is available as an open source software package at https://github.com/lujonathanh/BETS.

DYNOTEARS: Structure Learning from Time-Series Data

“…In tables 2 and 3, we compare the performance of DYNOTEARS to that of other methods. We obtain performance metrics for other algorithms from Lu et al (2019).…”

“…We preprocess our data and choose hyperparameters λ A and λ W through 10-fold cross validation, details of which can be found in Appendix D.1. In Lu et al (2019), the authors compared different approaches to learning these networks, including methods based on mutual information, Granger causality, dynamical systems, decision trees, Gaussian processes (GPs), and DBNs. Unsurprisingly, their results indicate that flexible nonparametric methods such as GPs and decision trees perform the best.…”

“…Unsurprisingly, their results indicate that flexible nonparametric methods such as GPs and decision trees perform the best. It is nonetheless instructive to compare the performance of DYNOTEARS to these methods for two reasons: 1) This gives us a sense of how much is lost in assuming the linear model (1), and 2) For an apples-to-apples comparison, we can still compare DYNOTEARS to the DBN methods (six in total) tested by Lu et al (2019) in AUROC (see Table 3 and Table 2, respectively). While the top performing methods were based on nonparametric models such as GPs, DYNOTEARS still outperforms several other nonparametric methods despite its use of the linear model (1).…”

Network inference with Granger causality ensembles on single-cell transcriptomics

“…Unlike other unsupervised aggregation approaches (Ahsen et al, 2019), SINGE's modified Borda counts do not assume that the ranked interaction lists are conditionally independent. Furthermore, SINGE's aggregation does not require generating a null distribution of A ij coefficients from permuted data (Lu et al, 2021), which is computationally more expensive but has the benefit of providing interaction false discovery rates.…”

“…Granger causality (Granger, 1969(Granger, , 1980) is a powerful approach for detecting specific types of causal relationships in long time series data. It has been used with bulk times series gene expression data (Fujita et al, 2010;Mukhopadhyay and Chatterjee, 2006;Shojaie and Michailidis, 2010;Finkle et al, 2018;Heerah et al, 2021;Lu et al, 2021), but these time series are typically short because of experimental limitations, making it more difficult to detect reliable gene-gene dependencies. The longer (pseudo)time series obtained from single-cell datasets make them appealing for Granger causality-based GRN reconstruction.…”