Improve the computation efficiency in epigenome-wide and genome-wide association studies

• Main Author: Tran, Nhat Sang
• Other Authors: Kwoh Chee Keong
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computer applications::Life and medical sciences
• Online Access: http://hdl.handle.net/10356/65735
• Institution: Nanyang Technological University

Background: Genome-wide association studies (GWAS) and epigenome-wide association studies (EWAS) hold the promise to explore the relationships among genetic variants, epigenetic changes and human diseases. The challenges lie in their computational burden due to the number of data returned from epigenetic measures (450k CpGs measured by Illumina Infinium 450k array) and genetic variants (millions of SNPs by sequence technology). As EWAS is a young and emerging topic, comprehensive computational supports are currently far behind the demands. An R package called GEM was created to discover how genetic variants (G) and environment factors (E) influenced methylation changes (M) in EWAS. The first generation of GEM uses linear model to determine the associations, so GEM finds it difficult to go through millions of regressions in large sample size. Solution: In this project, we implement the second generation GEM. We replaced the linear regression in the old GEM package with the newly developed semi-parallel approach. We first simulated pseudo methylation data, SNP data and environment data. Then we benchmark new Gmodel and Emodel by comparing the results with the standard respective functions in the old GEM. We showed the new Emodel can achieve around 500 times of efficiency with 1,000 samples and 10,000 CpGs; Gmodel can greatly improve the efficiency of more than 1,500 times with the same sample and CpG size and 60,000 SNPs. Conclusion: We implemented the new models and reported the computational efficiency of them. We also analysed the quality of accuracy in their results. This quality control process proved that our solution is reliable and should be applied in real study.