modSaRa: a computationally efficient R package for CNV identification

Abstract

Summary

Chromosomal copy number variation (CNV) refers to a polymorphism that a DNA segment presents deletion or duplication in the population. The computational algorithms developed to identify this type of variation are usually of high computational complexity. Here we present a user-friendly R package, modSaRa, designed to perform copy number variants identification. The package is developed based on a change-point based method with optimal computational complexity and desirable accuracy. The current version of modSaRa package is a comprehensive tool with integration of preprocessing steps and main CNV calling steps.

Availability and Implementation

modSaRa is an R package written in R, C ++ and Rcpp and is now freely available for download at http://c2s2.yale.edu/software/modSaRa.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Copy number variation refers to a polymorphism that a segment of DNA sequence presents deletion or duplication in the population. Modern technologies now increasingly allow accurate identification of copy number variants (CNVs) and many studies have explored the association between them and risk of complex diseases (Fanale et al., 2013; Poultney et al., 2013; Zahnleiter et al., 2013). Detection of CNVs has been proven essential in understanding the influence in human complex diseases biologically and epidemiologically. With the development of modern technologies, various complex data have been generated hence efficient statistical methods and computational tools are highly demanded. Methods based on change-points searching has been investigated for CNV analysis for a long time, including binary segmentation method (Vostrikova, 1981), circular binary segmentation (CBS) (Olshen et al., 2004), and exhaustive search methods (Braun et al., 2000; Yao, 1989). However, most of these popular methods limit their usage in modern big data analysis with computational complexity usually at least $O(n^2)$.

More recently, a screening and ranking algorithm (SaRa) was proposed with computational complexity$O(n)$ (Niu and Zhang, 2012). However practical issues arise from the application of this method in array-based data, including over-sensitivity and robustness of the normal based model. To address these issues, a modified SaRa method (modSaRa) was proposed as an endeavor to detect CNVs with optimal performance in practice (Xiao et al., 2015). Multiple bandwidths are used in change-point search to ensure better coverage. The candidate segments are clustered to CNVs via a Gaussian mixture model based clustering step. More details of the algorithm and applications implemented in modSaRa are described in the method paper (Xiao et al., 2015). A usable code of modSaRa was made available to the public.

Here, we present a comprehensive package of the modSaRa method. Compared to the originally release, this version includes useful new features as follows. We make use of Rcpp for a seamless integration of C ++ and R which greatly increases the computational speed. A user-friendly manual is embedded in the installation package that facilitates readers’ usage. The two preprocessing procedures are implemented for greater convenience than the original release. Importantly, CNVs are known to be important risk factors in cancer research. In this application note, we present an overview of the package and include a real dataset to illustrate the usage of modSaRa. The Additional file 1 contains a manual (the vignette for the package).

2 Methods and implementation

modSaRa formulates the analysis of array-based data as a problem of detecting change points. Let Ybe the data. A high dimensional normal mean based model was proposed.

$$Y_i={\mu }_i+{\epsilon }_i,\hspace{0.17em}{\epsilon }_i\sim N(0,{\sigma }^2),\hspace{0.17em}1\le i\le n$$ (1)

where μ is the vector for the underlying mean. A vector τ consists of change-point locations. The major task is to search for the number and the exact locations of change points. Next, modSaRa adopts screening and ranking algorithm with essential implementations to accurately identify change points, including Gaussian mixture model based clustering approach to filter false positives.

We then developed and integrated a set of functions in an R package called modSaRa. modSaRa is executable for both Windows and Linux. The package consists of two main components, preprocessing procedures and main CNV calling step (Supplementary Information). We implement the preprocessing steps in Rcpp (an R package to seamlessly integrate C ++ and R) which greatly improves the speed. The modSaRa package is freely available at http://c2s2.yale.edu/software/modSaRa.

3 Example usage

Two input files are required: a signal intensity file and a map file. The signal intensity file contains information for one marker per line. Each column represents data for one sequence. The map file provides information about marker name, chromosome ID and coordinates. Detailed instruction and output format can be found in the Additional file 1 and Supplementary information.

Figure 1 displays the plot of signal intensities of two CNVs identified by modSaRa with a melanoma study (Amos et al., 2011). Shown in the figure are two CNVs identified from two individuals on the same chromosomal region. LRRs usually cluster around zero when no duplications or deletions are present. In the identified CNV region, LRRs are obviously below zero, which indicates that these are deletions. Technically, the clustering pattern of BAFs also provides subtle information about the state. For deletions, they usually cluster around 0 and 1 (without cluster around 0.5). For the normal state, they usually congregate around 0, 0.5 and 1. Combining the information from the LRR and BAF distributions, the two CNVs were valid calling results.

Fig. 1.

CNVs identified by modSaRa from a melanoma study. The region between two vertical lines are identified CNVs. The horizontal axis indicates relative map locations on the chromosome and the vertical axis stands for LRR values (upper panel) and B allele frequency (BAF) (bottom panel)

4 Significance and conclusions

Recent advances in sequencing technology greatly reduce cost in next generation sequencing (NGS), and therefore, require more computationally efficient methods such as modSaRa. For NGS data, the problem becomes to find the break points along the sequences which are in read counts. Extension of the modSaRa to NGS data will be complicated yet plausible if the discreteness of the intensities is appropriately modeled, which warrants further development.