Mining gene regulatory networks by neural modeling of expression time-series

M. RUBIOLO, D.H. MILONE, G. STEGMAYER

Buenos Aires

Simposio; ASAI 2016; 2016

SADIO

Given the expression levels of a set of interacting genes measured at different time points, formal methods have been traditionally developed to model gene interactions. As a more recent alternative, the discovery of GRNs by data-driven methodologies has been under study in more recent years. In particular, artificial neural networks (ANN) can model pairs of genes activity over a number of time steps in order to infer genetic networks. Thus, all the possible combinations between genes must be analyzed in order to discover their relations. Using neural networks for this task  requires training them to predict a target gene regulation from candidate regulating profiles. By adjusting their synaptic weights, NNs alter their configuration to model each gene connection, which results in a minimum error in predicting a target profile.This work proposes a novel approach to discover a GRN from temporal genes expression profiles by using a pool of  ANNs with temporal delays at the input, named GRNNminer. Each NN is designed to discover, for a target gene profile at the output, the potential regulator of that gene at the  input,  by  modeling  their gene-to-gene interaction  during  a  time  period.