Genome-wide Association Study And Machine Learning Integrated Analysis Of Multimodal Data In Systemic Lupus Erythematosus

Objectives:To explore the genetic structure of Han systemic lupus erythematosus(SLE)patients in Southwest China and search for SLE-associated risk single nucleotide polymorphisms(SNPs)in this population.To explore the association between SLE risk SNPs alleles and multimodal data,such as organ involvement,disease activity,and brain magnetic resonance imaging,etc.Methods:A genome-wide association study(GWAS)was performed on 723 Han SLE patients and 2287 healthy controls in Southwest China,and significant genome-wide loci and their lead SNPs were screened and related genes were annotated.Based on the GWAS of samples from Southwest China and the latest GWAS data from other regions of China,the meta-analysis was performed to find the genome-wide significant loci and their lead SNPs in SLE.Based on the latest GWAS data of SLE populations in other regions of China,the polygenic risk score was calculated on samples from Southwest China.In SLE patients,the alleles of SLE risk lead SNPs screened out from GWAS and meta-analysis were used as groups to compare the differences in systemic involvement and laboratory test data between patients with SNP risk alleles and patients with SNP non-risk alleles.The classification and prediction of patients with different disease activity based on SLE risk SNP group network through deep learning were proceeded.The differences in brain structure and functional imaging between SLE risk allele carriers and non-risk allele carriers in SLE patients and healthy controls were compared,and the association between the SLE risk SNP group network and brain network was explored using machine learning.Results:1.Two genome-wide significantly SLE-associated loci were found based on GWAS of the population in Southwest China.The lead SNPs of these loci were rs145543103 and rs117026326,of which rs145543103 was located in the intergenic region of HLA-DRA and HLA-DRB5,and rs117026326 was located in the intron region of GTF2I.2.Meta-analysis based on the GWAS of the southwestern Chinese population and the GWAS of the northern,central,and southern populations in China found 42 genome-wide significant loci,of which the lead SNPs of the novel loci were rs72644150,rs373236624,rs210140,rs28626234,rs73622141,rs2970090,rs77992511,which are located in the intron region of MIR181A1HG,the intron region of WDR46,the intron region of BAK1,the intergenic region of PRDM1 and ATG5,the intron region of IKBKB,the intergenic region of IRF8 and LINC01082,respectively.3.SLE-associated genes were mainly enriched in the spleen,blood,small intestine,and lung.4.Based on the SLE GWAS data of the northern,central,and southern populations of China,a PRS model was established using the p-value clumping+thresholding method to predict SLE in the southwestern population of China.The p-value threshold was 5×10-8,which explained about 1.75%of the variation in the southwestern population.While the best model obtained by the Lasso sum method explained 25.6%of the variation in the southwest population samples.5.There were 5 SLE risk SNPs associated with systemic involvement in SLE patients.The risk alleles of rs548234 and rs2970090 were found to be protective factors for blood system involvement in SLE patients.And rs145543103 risk allele carrier was an independent risk factor for subclinical/clinical hyperthyroidism phenotype,while rs7579944 risk allele was an independent risk factor for neurological involvement phenotype in SLE patients.Patients carrying the rs72644150 risk allele had lower SLEDAI.6.The risk alleles of rs11185603,rs2362475,rs780669,rs7579944,rs145543103,rs210140,rs548234,rs73622141,rs117026326,rs28626234,rs2970090,rs6964608 and rs72644150 were associated with many clinical indexes,include blood cells counts,renal and hepatic function,immunoglobulins,tumor markers,thyroid function,coagulation routine,lymphocyte subsets,and cytokines,respectively.7.The LSTM model based on the SNP group network can achieve a prediction accuracy of 0.8171 for distinguishing SLE patients with active disease and inactive disease.8.The risk alleles of rs11185603,rs2362475,rs780669,rs7579944,rs145543103,rs210140,rs548234,rs73622141,rs117026326,rs28626234,rs2970090,rs6964608,rs72644150 were associated with changes of brain structure and function in SLE patients.9.There were differences in the multimodal brain network between SLE patients and healthy controls(HCs),and there were changes in brain structure and function in SLE patients compared with HCs.10.There was an association between the characteristics of the SNP group network and the brain network in the subjects.The association between the SNP group network and the brain network in SLE patients was not significantly different from that of the HCs,but 14 of the 16 SLE risk SNPs showed different weights in the association of SNP group network and brain network between SLE patients and HCs.Conclusions:1.SLE patients in southwest China presented different genetic structures compared with patients from the northern,southern,and central regions of China.This study found 2 genome-wide significant loci and one of them was novel.The meta-analysis found 42 genome-wide significant loci,of which 7 were novel.All of the novel loci were found to be involved in multiple immune processes,furthermore,lead SNPs of these loci were associated with multiple clinical manifestations.2.The LSTM model based on the SNP group network could classify patients with different SLEDAI as well as predict their disease activity.SNP group network was also associated with the structural and functional brain network of SLE patients.