Gene Association Analysis Of Generalized Linear Model Under Genetic Model Uncertainty

Gene analysis methods are of great significance to the research and treatment of genetic diseases.General gene analysis uses genome-wide association analysis,which can analyze genes in diseased individuals.Genome-wide association analysis usually considers that the genetic model of an individual's disease is known,but in practical problems,it is often impossible to determine the genetic model that affects the disease.If a wrong genetic model is specified,it may lead to the statistical power of the model.reduce.Secondly,not all genetic loci have an impact on individual disease,that is,high-dimensional genetic data is sparse,so variable selection needs to be performed while building a genetic model to screen out effective genetic loci.In addition,in addition to genetic factors,it also includes the influence of non-genetic factors such as environment and gender on the individual,and the influence of this nonlinear part can be considered by fitting the B-spline curve.The gene association analysis of the generalized linear model under the uncertain genetic model given in this paper firstly obtains three gene models under the uncertainty of the genetic model,namely the dominant,recessive and additive models.It is observed that the coefficients of these three gene models exist.At the same time,in order to solve the multi-collinearity existing in the model,a generalized additive model with uncertain genetic model is established,and because the phenotypic variable of an individual's disease is a binary variable with values of0 and 1,so A generalized linear logistic model for multiple loci was established.In order to screen out effective gene loci,two variable selection methods,LASSO and SCAD,are used here.This method achieves a sparse effect by compressing the model coefficients.The obtained model variable screening accuracy is improved compared with the original model,and with With the increase of the number of SNPs,the accuracy of model screening and the fitting effect of the model are better.