Complex disease gene cooperation and association analysis method based on GPU parallel computing

Abstract

The invention provides a complex disease gene cooperation and association analysis method based on GPU parallel computing. The method comprises the following steps of 1, performing main effect and epistasis detecting algorithm researching based on background control, namely utilizing a two-period strategy: in a first period, in all main effects and epistasis effects, deleting independent variables, and keeping relatively small number of variables; in a second period, inputting the variables which are screened in the first period into the model for finally determining a variable selecting method; 2, establishing an acceleration algorithm based on GPU parallel computing, and obtaining an MRMLM-GPU new method; and 3, performing performance analysis of the MRMLM-GPU new method. The method utilizes epistasis expansion for performing background control on the MRMLM algorithm and GPU acceleration improvement. In the epistasis primary screening period of the MRMLM algorithm, the SNP and the SNP pair are independent; and PyCUDA supplies a simpler programming mode for realizing parallel operation of the GPU. The method supplies a feasible new method for predicting the interaction genes of human diseases.

Description

technical field [0001] The invention relates to the technical field of medical statistics, in particular to a GPU-based parallel computing method for complex disease gene interaction correlation analysis. Background technique [0002] Genome-wide association study (GWAS) has successfully revealed the association of tens of thousands of single-nucleotide polymorphisms (single-nucleotide polymorphisms, SNPs) with complex disease traits. Currently, high-throughput techniques are capable of measuring up to ~5 million SNPs per individual, which can be more than doubled by imputing unmeasured SNPs from reference data. Genome-wide assays that detect one SNP at a time focus on the marginal effects of individual loci, however, for most common diseases, almost all associated SNPs have small effects and collectively explain the absence of phenotypic heritable variation. part. This phenomenon often creates the "missing heritability" problem. Although many factors such as genetic vari...

AI Technical Summary

Problems solved by technology

[0011] The technical problem to be solved by the present invention is to provide a complex disease gene interaction correlation analysis method based on GPU parallel computing, which is an improved algorithm based on multi-locus random SNP effect mixed linear model MRMLM, which extends the model to main effect Epistasis model, parallelize the algorithm, and use GPU to accelerate, so as to make up for and improve the deficiency of the existing continuous phenotype epistasis detection method in human genetics, and greatly improve the calculation while maintaining high efficiency and low false positive rate efficiency

Images