README.md

In macintyrelab/CNpare: Comparing tumour copy number profiles

CNpare

Summary of the package

CNpare is designed to compare and contrast genome-wide tumour copy number profiles. CNpare can be used for:

• Calculating similarities between copy-number profiles
• Quantifying the percentage of genome differences between copy-number profiles
• Comparing copy-number profiles based on copy-number signatures

Installation

You can install the development version of CNpare as follows:

git clone https://github.com/macintyrelab/CNpare.git

Once installed, CNpare must be loaded

library(CNpare)

Workflow overview

Input data

The most basic initial input to the CNpare package consists of a data frame containing the segmented copy number profile from a tumor sample.

This structure must contain the following columns:

• chromosome – chromosome
• start position – start
• end position – end
• copy number – segVal
• sample identifier – sample

Alternatively, a QDNAseqCopyNumbers object can be also used as input.

The segmented profile used as input needs to be transformed to bin-level copy number profiles. This can be done using getCNbins(), which needs the following arguments:

• posBins – genomic positions of bins. These can be obtained using getBinsStartsEnds()
• data – segmented copy number
• samples – vector with the name of samples

Package data

This package includes cells_segcn.RData, a dataframe with copy-number profiles of 1,417 human cancer cell lines from the Cancer Cell Line Encyclopaedia (CCLE) and Genomics of Drug Sensitivity in cancer (GDSC), which has been generated with ASCAT.sc. The original file can be found in the following github repository: VanLoo-lab/ASCAT.sc (https://github.com/VanLoo-lab/ASCAT.sc). This dataset must also be transformed to bin-level copy numbers before to be used for comparisons.

Compare two copy-number profiles

The function getSimilarities() can be used to get all similarity metrics calculated by CNpare. This function needs at least two bin-level copy number profiles to compare.

Four similarity metrics are computed for comparison by default, but the user can be interested in only computing one of them. It can be done by modifying the method parameter as follows:

• Pearson correlation – method="pearson"
• Manhattan distance – method="manhattan"
• Euclidean distance – method="euclidean"
• Cosine similarity – method="cosine"

The user can choose to compute the empirical p-value of each comparison by adding the argument pvalue=TRUE. To calculate the empirical p-value, CNpare uses the distribution of similarities obtained after comparing all samples included in the CNpare dataset. The function plot_simdensity can be used to visually inspect the uniqueness of the similarity value respect to the observed values.

Compute the % genome differences

Using the function getDifference(), the extent of genome differences (%) between two profiles can be calculated. This function provides the user the option to calculate the % genome difference normalized by ploidy status by adding the argument method="normalized"

Cluster profiles based on their exposure to copy-number signature

First, CNpare extracts the activity levels of the 7 ovarian copy-number signatures using the computational approach developed by Macintyre el al. Nature Genetics, 2018. Functions used for this purpose are:

• extractCopynumberFeatures()
• generateSampleByComponentMatrix()
• quantifySignatures()

Data needed for quantifying these signatures is included in CNpare, except for component_parameters.rds which may be downloaded from here: https://bitbucket.org/britroc/cnsignatures/src/master/data/. We are preparing an experiment data package for including all data needed for identifying copy-number signatures

The getClusterSamples() function can be used to get names of samples included in the same (include=TRUE, default) or closest (include=FALSE) cluster to the sample of interest according to the exposure to each copy number signature. This function automatically identifies the optimal number of clusters. Cosine similarity is used as distance metric for clustering.

Finally, using the function plotClusters(), samples are plotted by showing the two signatures with the highest variation across cluster means. Clusters are also represented.

Manuscript analysis Rmarkdown

The CNpare_Workflow rmarkdown in the /vignettes folder can be compiled to reproduce a typical workflow of the tool.

The Rmarkdown documents in the macintyrelab/CNpare_analyses repository (https://github.com/macintyrelab/CNpare_analyses) can be compiled to reproduce the analyses performed in the manuscript.

To compile these documents run knitr to html. Packages required to compile Rmarkdown documents include: knitr, dplyr, ggplot2, reshape2, data.table, flexmix, NMF, stats, splitstackshape, kableExtra, qusage, string and magrittr.

macintyrelab/CNpare documentation built on April 15, 2022, 4:46 a.m.