Genetic regulatory network inference using Recurrent Neural Networks trained by a multi agent system

We propose a novel algorithm for gene regulatory network inference. Gene Regulatory Network (GRN) inference is approximating the combined effect of different genes in a specific genome data. GRNs are nonlinear, dynamic and noisy. Time-series data has been frequently used for GRN modeling. Due to the function approximation and feedback nature of GRN, a Recurrent Neural Network (RNN) model is used. RNN training is a complicated task. We propose a multi agent system for RNN training. The agents of the proposed multi agent system trainer are separate swarms of particles building up a multi population Particle Swarm Optimization (PSO) algorithm. We compare the proposed algorithm with a similar algorithm that uses RNN with standard PSO for training. The results show improvements using the E. coli SOS dataset.