Reverse engineering highlights potential principles of large gene regulatory network design and learning

Overall organization of the FRANK algorithm (Fast Randomizing Algorithm for Network Knowledge). a Network graph for four genes (two transcription factors [TF]) and two targets (TA). Because of its simple architecture FRANK is able to simulate very large networks (accepting several thousands of TF and TA) and associated gene expression. The model accepts positive, negative and auto-regulations of TFs. b The Network graph in panel A is formalized as a network matrix named N made of two sub-matrices: A (TF effect on each others) and B (TF effect on TA). Gene expression at step 0 is then randomized (E 0) made of two sub-vectors: V 0 (being the expression values of TF) and W 0 (being the expression values of TA). c Formulas used to iteratively simulate gene expressions across iterative “time points” t. Gaussian noise simulating experimental transcriptomic measurements is adde