Fig. 3
From: Learning dynamical information from static protein and sequencing data
Reconstructed MEP networks for multistable gene-regulatory networks. We compare predicted MFPTs with direct estimates from stochastic simulations (Methods). a We performed simulations of three repressilator-type gene-regulatory network motifs with self-activation (left), consisting of two (top), three (middle), and four (bottom) genes, denoted A, B, C, and D. In the stochastic simulations (right), the numbers of each protein fluctuate between metastable states, but deterministic simulations of the system of ordinary differential equations (ODEs) at low molecule numbers are not able to identify the states, instead converging to a single steady state where all protein numbers are equal. b Disconnectivity graph and illustration of the identified lowest energy state for the four-dimensional system. In the lowest energy state (State 1), larger numbers of proteins a and c are present, leading to the activation of their associated genes and repressing the expression of proteins b and d. The next lowest energy state (State 2) is the converse scenario, with large numbers of proteins b and d present. c Predicted MFPTs agree well with those calculated from the time-dependent stochastic simulations for all three of the network motifs.
