GAIA: Genomic Analysis of Important
Aberrations
Sandro Morganella
Stefano Maria Pagnotta
Michele Ceccarelli
Contents
1
Overview
1
2
Installation
2
3
Package Dependencies
2
4
Vega Data Description
2
4.1
Marker Descriptor Matrix . . . . . . . . . . . . . . . . . . . . . .
2
4.2
Aberrant Region Descriptor Matrix . . . . . . . . . . . . . . . . .
3
4.3
Real aCGH Datset . . . . . . . . . . . . . . . . . . . . . . . . . .
3
4.3.1
Colorectal Cancer (CRC) Dataset
. . . . . . . . . . . . .
4
5
Function Description
4
5.1
load_markers . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
5.2
load_cnv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
5.3
runGAIA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
6
Homogeneous Peel-off
6
7
Run GAIA with Approximation
7
8
Integration of GAIA and Integrative Genomics Viewer
7
1
Overview
A current challenge in biology is the characterization of genetic mutations that
occur as response to a particular disease. Development of array comparative
genomic hybridization (aCGH) technology has been a very important step in
genomic mutation analysis, indeed, it enables copy number measurement in hun-
dreds of thousands of genomic points (called markers or probes). Despite the
high resolution of aCGH, accurate analysis of these data is yet a challenge. In
particular a major difficulty in mutation identification is the distinction between
driver mutations (that play a fundamental role in cancer progression) and pas-
senger mutations (which are random alterations with no selective advantages).
This document describes classes and functions of GAIA (Genomic Analysis
of Important Aberrations) package. GAIA uses a statistical framework based
1
on a conservative permutation test allowing the estimation of the probability
distribution of the contemporary mutations expected for non-driver markers.
Afterwards, the computed probability distribution and the observed data are
used to assess the statistical significance (in terms of q-value) of each marker.
Finally an iterative “peel-off” procedure is used to identify the most significant
independent regions which have a high probability to correspond to driver mu-
tations.
GAIA algorithm is carefully described in [1].
2
Installation
Install GAIA on your computer by using the following command:
source("http://bioconductor.org/biocLite.R")
biocLite("gaia")
GAIA package can be loaded in R by using the following command:
> library(gaia)
3
Package Dependencies
In order to use GAIA you need of the R package qvalue available at the bio-
conductor repository.
Note that by using the installation command biocLite the qvalue dependency
is automatically fulfilled. In contrast if GAIA has been manually installed (e.g.
using R CMD INSTALL command) you can install qvalue package by the follow-
ing command:
source("http://bioconductor.org/biocLite.R")
biocLite("qvalue")
4
Vega Data Description
In this section the data object used by GAIA will be described. This description
will be supported by the data provided in GAIA package.
4.1
Marker Descriptor Matrix
This matrix contains all needed informations about the observed markers (also
called probes). This matrix has a row for each marker and each marker is
described by a set of column reporting the name of the probe and its genomic
position. In particular the matrix has the following columns:
Probe Name : The name of the observed probe;
Chromosome : The chromosome where the probe is located;
Start : The genomic position (in bp) where the probe starts;
2
End : The genomic position (in bp) where the probe ends;
The column specifying the End position is optional and when it is missed,
start and end positions are considered to be coincident. This is the case of the
data provided within the package.
In order to load the marker descriptor matrix provided in GAIA use the
following command:
> data(synthMarkers_Matrix)
This marker descriptor matrix simulates the genomic position for the probes
of 24 chromosomes (each chromosome has 1000 probes). Chromosomes 23 and
24 represents sex chromosomes X and Y respectively.
4.2
Aberrant Region Descriptor Matrix
This matrix contains all needed informations about the observed aberrant re-
gions.This matrix has a row for each aberrant region and each of them is de-
scribed by the following columns:
Sample Name : The name of the sample in which the aberrant region is
observed;
Chromosome : The chromosome where the aberrant region is located;
Start : The genomic position (in bp) where the aberrant region starts;
End : The genomic position (in bp) where the aberrant region ends;
Num of Markers : The number of markers contained in the aberrant region;
Aberration : An integer indicating the kind of the mutation.
The column Aberration can assume only integer values in the range 0, · · · , K −
1 where K is the number of considered aberrations. For example if we are
considering loss, LOH and gain mutations than the only valid values for the
column Aberration are 0, 1 and 2.
In order to load the aberrant region descriptor matrix use the following
command:
> data(synthCNV_Matrix)
This aberrant region descriptor matrix simulates 10 samples for the chromo-
somes described by synthMarkers_Matrix. A summary of the aberrant regions
contained in this matrix follows:
4.3
Real aCGH Datset
In GAIA a real aCGH dataset is also provided, in the next we provide a brief
description of this dataset. Raw data were preprocessed by PennCNV tool [5]
and segmented by using Vega R/Bioconductor package [4].
3
Chromosome
Frequency Percentage
Start
End
Aberration Kind
1
100%
301
700
0
2
80%
301
700
0
3
60%
301
700
0
4
40%
301
700
0
5
20%
301
700
0
10
100%
1
700
1
11
80%
1
700
1
12
60%
1
700
1
13
40%
1
700
1
14
20%
1
700
1
20
100%
801
1000
2
21
80%
801
1000
2
22
60%
801
1000
2
23
40%
801
1000
2
24
20%
801
1000
2
Table 1: synthMarkers_Matrix: Aberrant Region Detail. The column Fre-
quency Percentage reports the percentage of samples containing the indicated
aberration.
4.3.1
Colorectal Cancer (CRC) Dataset
CRC dataset is composed by 30 samples hybridized on SNP 250k Affymetrix
GeneChip array [2]. Raw data are available in GEO with identifier GSE13429.
Patients of this dataset were diagnosed with microsatellite-stable CRC without
polyposis.
In order to run GAIA on this dataset both aberrant region descriptor matrix
(crc) and marker descriptor matrix (crc_markers) are provided.
5
Function Description
In this section all exported functions of GAIA package are described. GAIA
works on aCGH data in which hundreds of thousands probes are contemporary
observed, so the data loading phase can be very time consuming. For this reason
in GAIA two data loading functions are provided (see Sections 5.1 and 5.2). By
using this functions the users can load just one time the data and save they into
data objects that can be used for several GAIA executions.
5.1
load_markers
This function builds the marker descriptor object used in GAIA from a marker
matrix as described in Section 4.1:
> markers_obj <- load_markers(synthMarkers_Matrix)
The marker_obj provides an easy access to the data informations contained
in the marker matrix and and it is organized as a list:
the element marker_obj[[i]] contains the descriptions of all observed mark-
ers for the i-th chromosome and it is organized as a matrix of dimension 2 × M
4
(M is the number of observed probes for the i-th chromosome). First and second
row of this matrix contains start and end position respectively.
5.2
load_cnv
This function builds the aberrant region descriptor object used in GAIA by
using both the region matrix described in Section 4.2 and the marker descriptor
object created by the function load_markers, in addition the number of the
analyzed samples must be passed as argument:
> cnv_obj <- load_cnv(synthCNV_Matrix, markers_obj, 10)
The cnv_obj is organized as a double list:
the element cnv_obj[[j]][[i]] contains the informations about the j-th
aberration of the i-th chromosome. In particular cnv_obj[[j]][[i]] is a ma-
trix of dimension N × M where N is the number of samples (in the example 10)
and M the the number of observed probes for the i-th chromosome. The element
cnv_obj[[j]][[i]][n,m] is equal to 1 if in the marker m of the chromosome
i a mutation of kind j is observed, it is equal to 0 otherwise.
5.3
runGAIA
This is the core function of the package, indeed it allows to execute GAIA algo-
rithm. In particular runGAIA has the following header:
runGAIA(cnv_obj, markers_obj, output_file_name="", aberrations = -1,
chromosomes = -1, num_iterations = 10, threshold = 0.25)
cnv_obj : The aberrant region descriptor object created by using the function
load_cnv (see Section 5.2);
markers_obj : The marker descriptor object created by the function
load_markers (see Section 5.1);
output_file_name : (default “”) The file name used to save the significant
aberrant regions. If this argument is not specified the results will be not
saved into a file;
aberrations : (default −1) The aberrations that will be analyzed. The default
value −1 indicates that all aberration will be analyzed;
chromosomes : (default −1) The chromosomes that will be to analyzed. The
default value −1 indicates that all chromosomes will be analyzed;
num_iterations : (default 10) if the number of permutation steps (if approxi-
mation is equal to -1) - the number of column of the approximation matrix
(if approximation is different to -1);
threshold : (default 0.25) Markers having q-values lower than this threshold
are labeled as significantly aberrant. This parameter must be a real num-
ber in the range 0 − 1;
5
hom_threshold : (default 0.12) Threshold used in homogeneous peel-off (for
more detail see Section 6). This parameter must be a real number in the
range 0 − 1 and by using values lower than −1 homogeneous peel-off is
disabled and standard peel-off procedure is used;
approximation : (default=FALSE) if approximation is FALSE then GAIA ex-
plicitly performs the permutations, if it is TRUE then GAIA uses an
approximated approach to compute the null distribution.
Suppose that we want to analyze all aberrations and all chromosomes and
that we want to save the results within the file CompleteResults.txt, than we
use the following command:
> results <- runGAIA(cnv_obj, markers_obj, "CompleteResults.txt")
Both in the file CompleteResults.txt and in the matrix variable results we
can found all significant aberrant regions. In particular the following column
are used to describe each significant aberrant region:
Chromosome : The chromosome where the aberrant region is located;
Aberration Kind : An integer indicating the kind of the mutation;
Region Start [bp ]: The genomic position (in bp) where the aberrant region
starts;
Region End [bp ]: The genomic position (in bp) where the aberrant region
ends;
Region Size [bp ]: The size (in bp) of the aberrant region;
q-value : The estimated q-value for the aberrant region.
So with the following command we print the significant aberrant regions
having the minimum q-values:
> results[which(results[,6]==min(results[,6])),]
Suppose now that we want to analyze only the aberration 1 on the chro-
mosomes 10, 11 and 14 and that we want to save the results within the file
Results.txt. In addition we increase the significance threshold value from 0.25
to 0.5, than we use the following command:
> results <- runGAIA(cnv_obj, markers_obj, "Results.txt", aberrations=1, chromosomes=c(10, 11, 14), threshold=0.5)
6
Homogeneous Peel-off
In GAIA a new peel-off procedure, called homogeneous, is available. By usinh
homogeneous peel-off significant regions are detected by using both statistical
significance and within-sample homogeneity, so that more accurate results can
be obtained. Homogeneous peel-off is disabled for values lower than 0.
Homogeneous peel-off can be used only if two kinds of aberrations are ob-
served. In the next we show as homogeneous peel-off can be used on CRC
dataset. The first step is composed by the loading of the data and the creation
of the respective markers and aberrant regions descriptor objects:
6
> data(crc_markers)
> data(crc)
> crc_markers_obj <- load_markers(crc_markers)
> crc_cnv_obj <- load_cnv(crc, crc_markers_obj, 30)
Now we can run GAIA with homogeneous pee-off on the chromosome 14 by
using the suggested value for hom_threshold of 0.12:
> res <- runGAIA(crc_cnv_obj, crc_markers_obj, "crcResults.txt",
chromosomes=14, hom_threshold=0.12)
Results of the computation are saved in the file crcResults.txt
7
Run GAIA with Approximation
In GAIA a new approach to compute the null distribution is available. This
approach performs an approximation of the permutations. In particular, after
computed the frequency of alteration for each sample θ
j
, GAIA simulates a
matrix with dimension N ×K (N is the number of samples and K is the number
of performed approximation) where each column is a vector in which the element
j has a probability for drawing 1 equal to θ
j
. This asymptotically approximates
the original simulation when M → ∞ and since M is large enough, it is well
approximated in practice. So we suggest to use this approach in high resolution
scenario.
With the following command we run GAIA in its approximated version on
all chromosomes by performing K = 5000 approximations.
> res <- runGAIA(crc_cnv_obj, crc_markers_obj, "crc_approx_Results.txt", hom_threshold=0.12, num_iterations=5000, approximation=TRUE)
Results of the computation are saved in the file crcResults.txt
8
Integration of GAIA and Integrative Genomics
Viewer
The Integrative Genomics Viewer (IGV) [6] is a high-performance visualization
tool for interactive exploration of large, integrated datasets. It supports a wide
variety of data types including sequence alignments, microarrays, and genomic
annotations.
From version 1.5 GAIA produces in output a “.gistic”file that can be loaded
in IGV so that computed q-values for all analyzed chromosomes can be plotted.
References
[1] Morganella,S. et al. (2010) Finding recurrent copy number alterations pre-
serving within-sample homogeneity. Bioinformatics, DOI: 10.1093/bioinfor-
matics/btr488.
[2] Venkatachalam,R. et al. (2010) Identification of candidate predisposing copy
number variants in familial and early-onset colorectal cancer patients. Int. J.
Cancer.
7
[3] Astolfi,A. et al. (2010) A molecular portrait of gastrointestinal stromal tu-
mors: an integrative analysis of gene expression profiling and high-resolution
genomic copy number. Laboratory Investigation 90(9), 1285-1294.
[4] Morganella,S. et al. (2010) VEGA: variational segmentation for copy number
detection. Bioinformatics 26(25), 3020-3027.
[5] Wang,K. et al. (2007) PennCNV: an integrated hidden Markov model de-
signed for high-resolution copy number variation detection in whole-genome
SNP genotyping data. Genome Research 17, 1665-1674.
[6] Integrative Genomics Viewer (IGV): http://www.broadinstitute.org/
software/igv/home
8
Document Outline
Dostları ilə paylaş: |