| The interactions among genes can be comprehensively analyzed and understood through gene regulatory networks to find genes that cause disease,which are crucial for the treatment of complex diseases such as tumor.So far,many scholars have proposed the joint inference algorithms based on the differential equation model for inference of gene regulatory network,but without analysis and discussion about the algorithms performance systematically.This subject stems from the project "research on modeling of high throughput DNA methylation across human tissues and prediction methods for disease risk"supported by National Natural Science Foundation of China.For the existing joint inference algorithms deducing gene regulatory network model,this paper analyzes and compares the anti-noise,complexity and convergence systematically.The main contributions are as following:(1)Data collection and processing.This paper uses a synthetic dataset,two real-world datasets and filters three datasets,where the simulation data was generated by E-CELL,the real-world datasets were collected from Stanford Microarray Database.(2)Collection the joint inference algorithm.For the gene regulation network,algorithms based on the differential equation model proposed by researchers are sorted out,and implement on the Matlab platform.(3)Influence factors and anti-noise performance analysis of the joint inference algorithms.By analyzing the factors that affect the performance of the joint inference algorithm and considering the effect of noise on the deduced gene regulation network,Genetic programming and particle filter was superior to other joint inference algorithms in anti-noise performance.(4)Complexity and convergence analysis of the joint inference algorithms.Firstly,the complexity and convergence of the joint inference algorithm are analyzed in terms of mathematical.Secondly,the synthetic data is used to verify.Finally,the results show that under the condition of Gaussian noise,Genetic programming and Kalman filter has the lowest complexity and the smallest estimated error,but Genetic programming and particle filter converges most quickly;under the condition of non-Gaussian noise,Genetic programming and particle filter has lower complexity and the best convergence performance.(5)Application of the joint inference algorithm in biomedicine.Genetic programming and particle filter deduced the gene regulation network using the yeast dataset and the human cell time-series dataset respectively,which can analyze and infer the relationship between the genes,together with the cause of the disease. |
