A block-based imputation approach with adaptive LD blocks for fast genotype imputation

This paper addresses the issue of improving long imputation time usually required for a large volume of SNP genotype data which can be easily obtained by biological experiments with the genomewide SNP chip or the next-generation sequencing technology. For this purpose, we propose a block-based imputation approach that generates adaptive LD blocks with observed SNP genotype data and applies an imputation procedure for each block separately. Also, we implemented the block based imputation to maximize the use of computing resources. Specifically, each task of block imputation is allocated to individual processor and is executed on each processor independently. Thus, multiple tasks of block imputation can be executed on multiple processors in parallel where the parallelization can reach up to the maximum number of processors allowed by user's computing environment. Our experiment was performed with Mao et al.'s prostate cancer dataset. The results show that our adaptive block approach can reduce the imputation time up to 60-70% of original imputation time given by MaCH without the loss of imputation accuracy.