Research On Reconstruction Of Gene Regulatory Networks Based On Evolutionary Algorithms

Gene regulatory networks?GRNs?play important roles in understanding the structure of cellular systems,studying biological dynamics,revealing complex regulatory relationships,and exploring the pathogenesis of diseases.Thus,gene regulatory network reconstruction has become the core issue in bioinformatics.In recent years,with the development of high throughput technologies,such as gene sequencing,DNA microarray,and proteomics,a large amount of gene expression data have been accumulated.Various models and methods have been extensively studied.This thesis,aiming at the problems of using fuzzy cognitive maps and recurrent neural networks to model gene regulatory networks,focuses on designing more efficient reconstruction algorithms from the perspective of the model learning methods based on evolutionary algorithms.The main works are summarized as follows:?1?A hybrid algorithm of multi-agent genetic algorithm and random forests based on fuzzy cognitive maps is proposed to reconstruct gene regulatory networks,termed as MAGARF_FCM-GRN.In the proposed algorithm,an operator based on random forests is designed to acquire gene ranking,which can help optimize the search space of solution in multi-agent grid.Meanwhile,three genetic operators of MAGA,including the neighborhood competition,the mutation,and the self-learning operators are improved to accelerate the convergence speed.The comparisons with existing learning algorithms show that MAGARF_FCM-GRN is efficient and accurate for reconstructing gene regulatory networks.?2?A two-step algorithm combining memetic algorithm and least absolute shrinkage and selection operator?LASSO?based on recurrent neural networks is proposed to reconstruct gene regulatory networks,termed as MALASSO_RNN-GRN.First,a local search operator is designed to improve the algorithm efficiency.Second,some parameters of recurrent neural network are learned by a memetic algorithm,thus,the problem can be transformed into a linear network reconstruction problem.Third,LASSO is used to learn the network structure.Finally,the effect of three parameters on MALASSO_RNN-GRN is discussed.The results show that MALASSO_RNN-GRN can solve the problem of large-scale gene regulatory network reconstruction in the category of algorithms based on recurrent neural networks.?3?A sparse and decomposed particle swarm optimization is proposed to reconstruct gene regulatory networks,termed as SDPSO_FCM-GRN,which is based on fuzzy cognitive maps.First,the fitness function is decomposed,and the problem is transformed into sub-problems.Second,the sparseness of networks is controlled,and the way of initialization and encoding of particle swarm optimization is improved.Third,the evolutionary process of particles is optimized.Finally,the standard particle swarm optimization and related benchmark algorithms are compared with SDPSO_FCM-GRN.The results show that SDPSO_FCM-GRN has excellent performance,and the reconstructed gene regulatory network is sparse.