Research On Detection Of DNA Copy Number Variation Based On Read Depth Method

With the rapid development of the next-generation sequencing,the cost of detection of copy number variations in the whole genome is greatly reduced and the efficiency greatly improved.Copy number variation has been identified as an important structural variation which widely exists in human,by interfering the gene expression,CNV can regulate the phenotype of human beings.Studies have shown that CNVs can affect many kinds of human disease's susceptibility and it also one of the important factors of some human disease.However,researchers have proposed various kinds of methods to detect CNV,but there are lots of shortcomings especially for the low coverage data,such as the relatively high false negative rate.Our paper are mainly detecting CNVs on human chromosome,including the following two points:(1)For existing CNV detection method based on low coverage data doesn't considering the noise and GC bias in the datasets,we proposed a novel wavelets and scale-space filtering based method for CNV detection.The method divides the raw read depth data into two parts,one is the raw read coverage data and the other is the noise,after GC correction and noise reducing and then detection of CNVs based on scale-space filtering and finally search the significant CNV regions.The experiments on real human chromosome show that our method can detect small CNV region and are better in detecting the types and regions of CNV.(2)For existing CNV detection method based on low coverage data has highly time complexity,we proposed a method called saraCNV to detect CNV based on a novel two-stage change-point based method.In the first stage,we utilize a screening and ranking algorithm to find the most probable change-points from read coverage data,thus forming a candidate set of change-point.In the second stage,we made a Gaussian convolution with the candidate change-point,by changing the size of the value of sigma,we present the candidate change-point in different scales and the find the significant CNV regions.Experiments on real data sets show that our method gain a relatively low FNR when compared with other low coverage data based method.