Design And Implementation Of Differential Genetic Analysis System Based On Reinforcement Learning

With the continuous development of gene sequencing technology and the declining price of gene sequencing,researchers can get more and more data about genes.It is of great significance to related researchers how to analyze these massive data and find more value from them.Aiming at the problem of "high dimensions and few samples" in analyzing gene expression data features in the field of differentially expressed genes,this dissertation proposed a feature dimension reduction method based on reinforcement learning,which reduces high-dimensional features to low-dimensional by combining reinforcement learning models.It can complete the selection of differentially expressed genes,and at the same time can solve the problem that the number of genes selected needs to be determined in advance in the traditional method.The main work of this dissertation is as follows:First,the gene chip data related to this subject were collected and the original data were preprocessed to obtain data format that can be processed by machine learning methods.That is the original gene expression matrix.Then,the statistical T-test algorithm combined with the improved Fold-change algorithm was used for preliminary feature gene selection.Aiming at the small difference in the average gene expression level,the FC value log2 conversion was used to significantly indicate the up-down relationship between different groups.the algorithm removed the genes whose gene expression value is very close and unrelated to grouping in the sample.And the characteristic genes were retained with quite different gene expression values in different groups.Secondly,the differential gene selection algorithm based on reinforcement learning was proposed,and a reinforcement learning model was constructed for high-dimensional small sample gene expression data characteristics.The entire learning process of the environment S was invariant,and the neural network was used as the actor.Two cases:"Selected" and "Unselected" was taken as actions,a improved reward function combined with the classification accuracy rate was given to select the most relevant differential genes.Finally,the system function modules were analyzed and designed,a differential gene analysis system was developed based on reinforcement learning,and the results were further visualized.Because the main audience of this system is professional researchers in biology and medicine,the graphics can provide concise and intuitive expressions for professional researchers in biology and medicine,which is helpful for theoretical demonstration and analysis of next experiments.The experimental and verification results show that the differential gene selection algorithm based on reinforcement learning has about 3%improvement in accuracy compared with the traditional algorithm.At the same time,by compared with medical data,this method can select out more correct genes,which has a strong effect on related medical research.The differential gene analysis system developed in this dissertation can met the actual needs of biomedical researchers and workers,and improve the efficiency of people analyzing gene chip data.