Limits on Inferring Gene Regulatory Networks Subjected to Different Noise Mechanisms

Abstract

One of the most difficult and pressing problems in computational cell biology is the inference of gene regulatory network structure from transcriptomic data. Benchmarking network inference methods on model organism datasets has yielded mixed results, in which the methods sometimes perform reasonably well and other times fail to outperform random guessing. In this paper, we analyze the feasibility of network inference under different noise conditions using stochastic simulations. We show that gene regulatory interactions with extrinsic noise appear to be more amenable to inference than those with only intrinsic noise, especially when the extrinsic noise causes the system to switch between distinct expression states. Furthermore, we analyze the problem of false positives between genes that have no direct interaction but share a common upstream regulator, and explore a strategy for distinguishing between these false positives and true interactions based on noise profiles of mRNA expression levels. Lastly, we derive mathematical formulas for the mRNA noise levels and correlation using moment analysis techniques, and show how these levels change as the mean mRNA expression level changes.