Leveraging Twin Information for Inference of Gene Regulatory Networks

Determining the architecture of gene regulatory networks is essential for understanding how cells process and respond to environmental stimuli. Single-cell transcriptomics provides the high-throughput data required for such inference en masse, but at the cost of losing dynamical information. Regulatory correlations can still be observed through intrinsic variation of transcriptional levels across isogenic single cells, albeit heterogeneity in cell states precludes accurate inference. Here, we overcome these challenges by developing TwINFER, a framework that utilizes information obtained from recently divided twin cells, identifiable with modern barcoding techniques. Incorporating twin correlations, we can discern between regulatory and non-regulatory correlations. Moreover, separating twins and measuring their transcriptome at different time points enables inference of the direction, type (activation/repression), and strength of regulatory interactions. As a proof-of-principle, we use an extensive set of simulations, covering common network motifs and large-scale networks. We then apply our framework to analyze real data.