Cellular-Level Gene Regulatory Networks: Their Derivation and Properties

Abstract

There is considerable direct and indirect evidence that gene regulatory networks are largely self-regulating, rather than exclusively regulated by upstream master control genes. This leads to several familiar properties of cells, such as homeostatic tendencies and robustness to internal and external noise. Moreover it implies that certain cell state transitions are only possible following the modification of the expression levels of many genes. Therefore these large collections of genes are functional units in the cell whose collective behavior can be modeled. This motivates the inference of gene regulatory networks, whose components are large groups of genes, a goal that by reducing the total amount of data required to infer the model, serendipitously allows the inference of regulatory networks inclusive of the entire genome. Highly predictive global regulatory networks can be derived using linear models applied to gene expression level temporal transitions, revealing a wealth of specific conclusions about cellular regulation at the highest levels. For example, ATP synthesis is shown to the predominant activator of large groups of genes, while gene groups involved in DNA replication repress transcription globally. Moreover, these relationships can be probed for their variance or invariance across scale by inferring networks composed of varying numbers of gene groups, equivalent to tuning the resolution of a map.