Copy Number Variation Detection On Multi-core Platform And Parallel Algorithm

It’s a powerful way to detect tumor gene copy number variations(CNVs) in studies when discovering cancer-causing genes. Developing of the next generation sequencing(NGS) technologies provides more data, which not only promote the algorithm for detecting CNVs, but also show great challenges in the efficiency during the data processing.In this paper, we constructed a set of CNVs detection processs based on the theory of Bayesian Information Criterion and the Read-depth method. In order to improving the precision of CNVs’ breakpoints, we designed an improved binning strategy in which the reads number in each bin is constant during the statistical process. The new binning strategy determined a set of unfixed initial bins dynamically according to the analyzed sequencing data, which improving the precision when predict the location of CNVs’ breakpoints.In this paper, we proposed a new kind of multi-core oriented parallel algorithms for the detection of CNVs, in order to solving the time-consuming problem when using large-scale sequencing data and achieving better detection performance.Then we did a feasiblility analysis for the parallelization after deep understanding the segmentation algorithms and realize the parallelization on the multicore platform by dividing the computing task into multiple BIC and BIC difference calculations. The test results show that our optimized algorithms can achieve 6.74 X speed up in the 8-core CPU platform against serial program.Because of the DNA specimen extracted from a mixture of tumor cells and normal stromal cells, we analyzed the process and principles of the absolute copy number estimation by the means of tumor purity and ploidy, and put the process of absolute copy number estimation and the foregoing content together to form a completive workflow in this paper.We analyzed the least squares objective function which is build by the relationship among tumor purity, ploidy and absolute copy numbers in each segment and improved the algorithm when solving the objective function.We paralleled the improved searching algorithm and did some experiments.The result show that the improvement and optimization of solving the objective function on multi-core platform largely enhance the absolute copy number detection performace.