However, physical HPC is costly and requires extensive maintenance.
The physical architecture of HPC consists of numerous processing units, large shared memory and huge data storage cooperatively functioning to obtain high performance usually measured as floating-point operations per second (FLOPS) ( Subramaniam & Feng, 2012). To overcome the bottleneck of data analysis, high performance computing (HPC) is now commonly used in large-scale bioinformatics tasks including sequence alignment ( Orobitg et al., 2015), simulation ( Zhang, Wong & Lightstone, 2014) and machine learning ( D’Angelo & Rampone, 2014). However, genomic and proteomic data are not readily usable or meaningful without proper analysis and interpretation which become the bottleneck of genomic and proteomic studies due to tremendous computational resource requirement ( Scholz, Lo & Chain, 2012 Berger, Peng & Singh, 2013 Neuhauser et al., 2013). Furthermore, metagenomics, a study of genetic materials in samples directly collected from particular environments, is now greatly advanced by high-throughput assays, and now becomes applicable to forensic sciences ( Fierer et al., 2010) and pathogen discovery ( Chiu, 2013). In precision medicine, clinicians can diagnose and tailor a treatment for a disease based on the patient profile derived from “omics” data ( Chen & Snyder, 2012). Breakthroughs in genomic and proteomic data generation lead to development and emergence of several disciplines. Up to 1 trillion bases can be sequenced in one 6-day run by Illumina HiSeq 2500 ( Rhoads & Au, 2015) while mass spectrometry can now completely analyse a proteome and quantify protein concentrations in an entire organism ( Ahrne et al., 2015). Massive data are now affordably, easily and frequently generated by genomic and proteomic assays such as massively parallel sequencing and high-throughput mass spectrometry. The grid implementation by BOINC also helped tap unused computing resources during the off-hours and could be easily modified for other available bioinformatics software. Thus, the grid implementation of BLAST by BOINC is an efficient alternative to the HPC for sequence alignment. The estimated durations of BLAST analysis for 4 million sequence reads on a desktop PC, HPC and the grid system were 568, 24 and 5 days, respectively. The result and processing time were compared to those from a single desktop computer and HPC. Sequencing results from Illumina platform were aligned to the human genome database by BLAST on the grid system. In order to test the performance of the grid system, we adapted the Basic Local Alignment Search Tools (BLAST) to the BOINC system. Fifty desktop computers were used for setting up a grid system during the off-hours. In this study, we implemented grid computing in a computer training center environment using Berkeley Open Infrastructure for Network Computing (BOINC) as a job distributor and data manager combining all desktop computers to virtualize the HPC. Other means were developed to allow researchers to acquire the power of HPC without a need to purchase and maintain one such as cloud computing services and grid computing system. However, the HPC is expensive and difficult to access. Due to the size of the data, a high performance computer (HPC) is required for the analysis and interpretation. Nevertheless, the data are not readily usable or meaningful until they are further analysed and interpreted. Consequently, a massive amount of experiment data is now rapidly generated. Development of high-throughput technologies, such as Next-generation sequencing, allows thousands of experiments to be performed simultaneously while reducing resource requirement.
0 kommentar(er)
