Inference of sparse gene regulatory network from RNA-Seq time series data

Inferring gene regulatory networks (GRNs) from transcriptomic data is a challenging problem in system biology. The advent of high-throughput sequencing technology like RNA-Seq has provided a more powerful platform to measure the gene expressions than the traditional microarrys. However, although many methods have been proposed to infer networks from microarray data, there is a need for the new methods capable of modeling the count-based and highly variable RNA-Seq data. The few recent RNA-Seq based methods have focused on the inference from non-temporal data, which only take snapshots of the system at one given time point. Time series data will provide a more thorough description of the physiological changes over time. In this paper, we proposed a novel algorithm for the inference of GRNs from the RNA-Seq time series data. We considered a log-linear model for the temporal evolution of the gene expressions, and modeled the gene expression levels by a negative binomial distribution. Furthermore, We added an 11 penalty to control the network sparseness. We evaluated the proposed algorithm using synthetic data obtained from randomly generated networks in different scenarios. In addition, we tested the algorithm on the in vivo time series data obtained during Nematostella vectensis development. The inferred network for the developing gut overlaps well with the same network derived from the fluorescent imaging of gene expressions.