366
V. Mittard-Runte et al.
classification phase, the trained classifier is applied to data and predicts class labels
for these.
One of the first applications of this method was the study by Golub et al. (1999)
who used microarray data for the classification of different types of leukaemia. One
of the most simple but highly useful methods is the k nearest-neighbour (kNN) classifier which makes class predictions for an unknown object based on the majority of
the k closest labelled examples (Cover and Hart 1967). Wu et al. (2005) have used
a kNN classifier to compare the merit of different normalization methods.
Vapnik (1999) developed the Support Vector Machine (SVM) approach for classification. SVMs are suited for finding an optimal separating function and allow
for linear and non-linear decision boundaries. Examples of applications of SVMs
to microarray data are frequently found in the literature (see for example, Pavlidis
et al. 2002, Liu et al. 2005) and they are becoming more and more popular.
9.4.2.6 Microarray Software
An almost unmanageable amount of software related to microarrays has been
developed over the recent years. The domains of application of these software
tools also often overlap. Here we will, therefore, concentrate on several examples.
Unfortunately, there is currently no perfect solution and all software packages have
their advantages and drawbacks. In particular there is no all-in-one solution that
offers both highly flexible analysis methods and an easy to use graphical user interface. For most analysis tasks, this means that the analysis will involve the use of
multiple software programs. Commercial software is available for most tasks, with a
large range of license fees as well as free open source solutions. Another important
aspect when choosing software is its hardware and software requirements. Some
software packages, especially database-based systems, require a significant effort
for installation and professional administration. Several categories of software are
described in the following sections.
9.4.2.7 Image Analysis Software
Image analysis software serves the purpose of analysing the resulting images
obtained from the scanner software. This process is usually twofold. A segmentation step is performed to identify the locations of the features and their boundaries
within the image, corresponding to spots of the microarray. Depending on the software, this step can be carried out manually, semi-automatically if predefined grid
information is used, or fully automatically.
The TIGR Spotfinder software for example, is an academic open-source project,
which is available for several operating systems. A package called Spot also exists.
This is implemented as a package for the statistical environment R. Users of
Affymetrix
R
microarrays have to use the MAS5 algorithm from Affymetrix
R
for
image analysis.
V. Mittard-Runte et al.
classification phase, the trained classifier is applied to data and predicts class labels
for these.
One of the first applications of this method was the study by Golub et al. (1999)
who used microarray data for the classification of different types of leukaemia. One
of the most simple but highly useful methods is the k nearest-neighbour (kNN) classifier which makes class predictions for an unknown object based on the majority of
the k closest labelled examples (Cover and Hart 1967). Wu et al. (2005) have used
a kNN classifier to compare the merit of different normalization methods.
Vapnik (1999) developed the Support Vector Machine (SVM) approach for classification. SVMs are suited for finding an optimal separating function and allow
for linear and non-linear decision boundaries. Examples of applications of SVMs
to microarray data are frequently found in the literature (see for example, Pavlidis
et al. 2002, Liu et al. 2005) and they are becoming more and more popular.
9.4.2.6 Microarray Software
An almost unmanageable amount of software related to microarrays has been
developed over the recent years. The domains of application of these software
tools also often overlap. Here we will, therefore, concentrate on several examples.
Unfortunately, there is currently no perfect solution and all software packages have
their advantages and drawbacks. In particular there is no all-in-one solution that
offers both highly flexible analysis methods and an easy to use graphical user interface. For most analysis tasks, this means that the analysis will involve the use of
multiple software programs. Commercial software is available for most tasks, with a
large range of license fees as well as free open source solutions. Another important
aspect when choosing software is its hardware and software requirements. Some
software packages, especially database-based systems, require a significant effort
for installation and professional administration. Several categories of software are
described in the following sections.
9.4.2.7 Image Analysis Software
Image analysis software serves the purpose of analysing the resulting images
obtained from the scanner software. This process is usually twofold. A segmentation step is performed to identify the locations of the features and their boundaries
within the image, corresponding to spots of the microarray. Depending on the software, this step can be carried out manually, semi-automatically if predefined grid
information is used, or fully automatically.
The TIGR Spotfinder software for example, is an academic open-source project,
which is available for several operating systems. A package called Spot also exists.
This is implemented as a package for the statistical environment R. Users of
Affymetrix
R
microarrays have to use the MAS5 algorithm from Affymetrix
R
for
image analysis.