Prognosis Of MCI Development To AD Based On Polygenic Risk Score And Feature Selection Method

Objective:Alzheimer’s disease(AD)is a progressive,irreversible neurodegenerative disease whose preclinical phase begins with mild cognitive impairment(MCI),and more than half of all MCI patients each year will develop AD within five years.In addition,APOE is an important gene for the occurrence of AD,and different treatments have different effects on the development of MCI to AD.Therefore,this study investigated the effect of different APOE treatments on the progression of MCI to AD based on genome-wide and candidate genomic data,and downscaled the data in terms of polygenic risk score(PRS)and feature selection to explore the best way to process the data during the progression of MCI to AD.Meanwhile,the cox proportional hazard model(CPH),random survival forest(RSF),and survival support vector machine(SSVM)were investigated to predict the prognosis of MCI to AD.The prognostic prediction performance of MCI to AD provides a stronger methodological theoretical basis for modeling the prognosis of MCI development to AD.Methods:Firstly,univariate analysis was conducted on the genetic and non-genetic factors related to AD,and the significant part was included as covariates in the following studies.Then,the two loci rs429358 and rs7412 corresponding to APOE were taken as covariates and included together with other SNPs after PRS calculation.The whole genome and candidate genome data were used to calculate PRS using Clumping and thresholding(C+T)and Polygenic risk score-Continuous shrinkage(PRS-CS),respectively,to aggregate numerous single nucleotide polymorphisms(SNPs)into a single score.At the same time,Least absolute shrinkage and selection operator(Lasso),Elastic-net(Enet)and Sure independence screening(SIS)methods were used for feature selection and dimensionality reduction of candidate genome data.The SNPs selected by feature selection methods were selected for simple path enrichment analysis using Metascape,an online enrichment analysis software.Then,the important SNPs obtained after feature selection were incorporated into the prediction model respectively,and the survival of the fifth year when MCI developed into AD was predicted and modeled by CPH,RSF and SSVM,three commonly used survival analysis methods.Finally,C-index is used as the evaluation index of model prediction effect.Results:This study ultimately retained 429 MCI individuals by controlling for ethnicity,outcome events,etc.After a five-year follow-up study,299 MCI patients remained,and 130 MCI individuals developed into AD.In univariate analysis,differences were found in the survival rates of MCI patients on genetic factors: whether they carried the APOE4 gene,and non-genetic factors: age,MMSE,Mo CA,CDR-SB,and ADAS11 scale scores,with Pvalues less than 0.05,which were statistically significant and were included as covariates in the next study.In calculating the PRS,it was found that the C-index of the candidate genome was higher than that of the whole genome in most cases,and only in the APOE.covariate model the C-index of the candidate genome was 0.007 lower than that of the whole genome.The C-index for both genome-wide and candidate loci containing APOE(whole genome:0.606,candidate: 0.625)was significantly higher than that of C-index without APOE(whole genome: 0.564,candidate: 0.617).And the C-index with two APOE loci as covariates was higher than calculating PRS with them,but the prediction was poorer after calculating PRS for two important loci of APOE under whole genome with C-index of 0.602.When downscaling the candidate genomic data using feature selection methods,the best prediction was achieved using the Enet method with a C-index of 0.809;the Lasso method with a Cindex of 0.796;and the SIS method with a C-index of 0.711.Second,model prediction based on SNPs screened by the Enet method showed that the CPH method for the fifth year of MCI development to AD The prognostic prediction was better than the RSF and SSVM methods,with a predicted C-index of 0.809 for CPH and 0.711 and 0.738 for RSF and SSVM,respectively,and the Welch test and multiple comparisons showed that the P-value of the three were less than 0.05,and the differences were statistically significant.Conclusion:Under the candidate genome,two loci of APOE were used as covariates,and the dimensionality reduction method using Enet for feature selection of the data was better than PRS,and the fifth-year survival prediction performance of MCI developing into AD after screening of Enet using CPH method was better than RSF and SSVM methods,therefore,this model can help physicians to identify patients at high risk of developing MCI into AD,and it provides a more feasible statistical modeling solution for its prognosis prediction.