Identification Of Alzheimer’s Biomarkers Based On Deep Neural Network And Gene Regulation Network Modeling And Noise Propagation

Alzheimer’s disease（AD）is a neurodegenerative disease for which there is cur-rently no complete cure.The accepted pathogenesis of AD is the amyloid cascade hypothesis and overexphosphorylation of theprotein.There is increasing evidence that AD is associated with changes in the expression of(6²⁺signaling disorder and(6²⁺signaling protein levels,with a positive feedback loop between-amyloid levels（）and calcium levels.Currently,many studies attempt to identify genetic risk factors from the perspective of single or multiple sets of data,but the accuracy and interpretability of these studies remain unsatisfactory.Deep learning is the latest breakthrough in data science that could be used to discover previously unknown features that can be learned directly from big data.In chapter 2,based on deep neural network Resnet18 model,we used attention mechanism module（SE module）to increase the sensitivity of Resnet18 to channel characteristics,using gene expression data of five human brain regions to classify AD patients and healthy controls with predictive accuracy up to 100%.Combining deep learning with bioinformatics,ten new biomarkers for identifying AD were obtained from differentially expressed genes in human brain regions:FBLN1,MAPK8IP2,KLC1,GRIN2A,SYT2,PVALB,CD44,EP300,DGAT1,SPTAN1.Further study of the KEGG pathway for these ten biomarkers revealed that GRIN2A and EP300are both on the long-term enhancer pathway in the hippocampus,and GRIN2A is important for the expression of(6²⁺and(6²⁺.Based on the AD(6²⁺hypothesis,GRIN2A promises to be a new target for early diagnosis of AD.Using the results of the second chapter,a differential equation involving,(6²⁺and GRIN2A is proposed based on the interaction mediated minimum ofand(6²⁺proposed by Jo（?）lle De Caluw’0)et al.Firstly,the linear noise approxi-mation of the main equation is used to study the model,and then the theoretical formula is simulated by Gillespie algorithm to study the relationship between noise and parameters.It was found that the noise number can be reduced by adjusting parameters,and the influence of noise is considered as the indirect cause of AD.Fur-ther,comparing the relationship between internal inherent noise and transmission noise,it is found that the source of noise in the network is mainly internal noise,but the effect of transmission noise on the network is also important.By regulating noise to keep the number of cells within normal range,so as to intervene with thedeposition and effectively control AD deterioration.