bga: Between group analysis
In made4: Multivariate analysis of microarray data using ADE4

Description Usage Arguments Details Value Author(s) References See Also Examples

Discrimination of samples using between group analysis as described by Culhane et al., 2002.

bga(dataset, classvec, type = "coa", ...)
## S3 method for class 'bga'
plot(x, axis1=1, axis2=2, arraycol=NULL, genecol="gray25", nlab=10, 
         genelabels= NULL, ...)

`dataset`	Training dataset. A `matrix`, `data.frame`, `ExpressionSet` or `marrayRaw-class`. If the input is gene expression data in a `matrix` or `data.frame`. The rows and columns are expected to contain the variables (genes) and cases (array samples) respectively.
`classvec`	A `factor` or `vector` which describes the classes in the training dataset.
`type`	Character, "coa", "pca" or "nsc" indicating which data transformation is required. The default value is type="coa".
`x`	An object of class `bga`. The output from `bga` or `bga.suppl`. It contains the projection coordinates from `bga`, the \$ls, \$co or \$li coordinates to be plotted.
`arraycol, genecol`	Character, colour of points on plot. If arraycol is NULL, arraycol will obtain a set of contrasting colours using `getcol`, for each classes of cases (microarray samples) on the array (case) plot. genecol is the colour of the points for each variable (genes) on gene plot.
`nlab`	Numeric. An integer indicating the number of variables (genes) at the end of axes to be labelled, on the gene plot.
`axis1`	Integer, the column number for the x-axis. The default is 1.
`axis2`	Integer, the column number for the y-axis, The default is 2.
`genelabels`	A vector of variables labels, if `genelabels=NULL` the row.names of input matrix `dataset` will be used.
`...`	further arguments passed to or from other methods.

bga performs a between group analysis on the input dataset. This function calls bca. The input format of the dataset is verified using isDataFrame.

Between group analysis is a supervised method for sample discrimination and class prediction. BGA is carried out by ordinating groups (sets of grouped microarray samples), that is, groups of samples are projected into a reduced dimensional space. This is most easily done using PCA or COA, of the group means. The choice of PCA, COA is defined by the parameter type.

The user must define microarray sample groupings in advance. These groupings are defined using the input classvec, which is a factor or vector.

Cross-validation and testing of bga results:

bga results should be validated using one leave out jack-knife cross-validation using bga.jackknife and by projecting a blind test datasets onto the bga axes using suppl. bga and suppl are combined in bga.suppl which requires input of both a training and test dataset. It is important to ensure that the selection of cases for a training and test set are not biased, and generally many cross-validations should be performed. The function randomiser can be used to randomise the selection of training and test samples.

Plotting and visualising bga results: 1D plots, show one axis only: 1D graphs can be plotted using between.graph and graph1D. between.graph is used for plotting the cases, and required both the co-ordinates of the cases (\$ls) and their centroids (\$li). It accepts an object bga. graph1D can be used to plot either cases (microarrays) or variables (genes) and only requires a vector of coordinates.

2D plots: Use plot.bga to plot results from bga. plot.bga calls the functions plotarrays to draw an xy plot of cases (\$ls). plotgenes, is used to draw an xy plot of the variables (genes). plotgenes, is used to draw an xy plot of the variables (genes).

3D plots: 3D graphs can be generated using do3D and html3D. html3D produces a web page in which a 3D plot can be interactively rotated, zoomed, and in which classes or groups of cases can be easily highlighted.

Analysis of the distribution of variance among axes:

It is important to know which cases (microarray samples) are discriminated by the axes. The number of axes or principal components from a bga will equal the number of classes - 1, that is length(levels(classvec))-1.

The distribution of variance among axes is described in the eigenvalues (\$eig) of the bga analysis. These can be visualised using a scree plot, using scatterutil.eigen as it done in plot.bga. It is also useful to visualise the principal components from a using a bga or principal components analysis dudi.pca, or correspondence analysis dudi.coa using a heatmap. In MADE4 the function heatplot will plot a heatmap with nicer default colours.

Extracting list of top variables (genes):

Use topgenes to get list of variables or cases at the ends of axes. It will return a list of the top n variables (by default n=5) at the positive, negative or both ends of an axes. sumstats can be used to return the angle (slope) and distance from the origin of a list of coordinates.

For more details see Culhane et al., 2002 and http://bioinf.ucd.ie/research/BGA.

A list with a class bga containing:

`ord`	Results of initial ordination. A list of class "dudi" (see `dudi` )
`bet`	Results of between group analysis. A list of class "dudi" (see `dudi`), "between" (see `bca`)
`fac`	The input classvec, the `factor` or `vector` which described the classes in the input dataset

Aedin Culhane

Culhane AC, et al., 2002 Between-group analysis of microarray data. Bioinformatics. 18(12):1600-8.

See Also bga, suppl, suppl.bga, bca, bga.jackknife

data(khan)

if (require(ade4, quiet = TRUE)) {
  khan.bga<-bga(khan$train, classvec=khan$train.classes)  
  }

khan.bga
plot(khan.bga, genelabels=khan$annotation$Symbol)

# Provide a view of the principal components (axes) of the bga  
heatplot(khan.bga$bet$ls, dend="none")

Loading required package: ade4
Loading required package: RColorBrewer
Loading required package: gplots

Attaching package: 'gplots'

The following object is masked from 'package:stats':

    lowess

Loading required package: scatterplot3d
$ord
$ord
Duality diagramm
class: coa dudi
$call: dudi.coa(df = data.tr, scannf = FALSE, nf = ord.nf)

$nf: 63 axis-components saved
$rank: 63
eigen values: 0.1713 0.1383 0.1032 0.05995 0.04965 ...
  vector length mode    content       
1 $cw    306    numeric column weights
2 $lw    64     numeric row weights   
3 $eig   63     numeric eigen values  

  data.frame nrow ncol content             
1 $tab       64   306  modified array      
2 $li        64   63   row coordinates     
3 $l1        64   63   row normed scores   
4 $co        306  63   column coordinates  
5 $c1        306  63   column normed scores
other elements: N 

$fac
 [1] EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS   
[11] EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS   
[21] EWS    EWS    EWS    BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL
[31] BL-NHL NB     NB     NB     NB     NB     NB     NB     NB     NB    
[41] NB     NB     NB     RMS    RMS    RMS    RMS    RMS    RMS    RMS   
[51] RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS   
[61] RMS    RMS    RMS    RMS   
Levels: EWS BL-NHL NB RMS

attr(,"class")
[1] "coa" "ord"

$bet
Between analysis
call: bca.dudi(x = data.ord$ord, fac = classvec, scannf = FALSE, nf = nclasses - 
    1)
class: between dudi 

$nf (axis saved) : 3
$rank:  3
$ratio:  0.3599779

eigen values: 0.1522 0.1218 0.08981

  vector length mode    content      
1 $eig   3      numeric eigen values 
2 $lw    4      numeric group weigths
3 $cw    306    numeric col weigths  

  data.frame nrow ncol content                       
1 $tab       4    306  array class-variables         
2 $li        4    3    class coordinates             
3 $l1        4    3    class normed scores           
4 $co        306  3    column coordinates            
5 $c1        306  3    column normed scores          
6 $ls        64   3    row coordinates               
7 $as        63   3    inertia axis onto between axis


$fac
 [1] EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS   
[11] EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS    EWS   
[21] EWS    EWS    EWS    BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL BL-NHL
[31] BL-NHL NB     NB     NB     NB     NB     NB     NB     NB     NB    
[41] NB     NB     NB     RMS    RMS    RMS    RMS    RMS    RMS    RMS   
[51] RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS    RMS   
[61] RMS    RMS    RMS    RMS   
Levels: EWS BL-NHL NB RMS

attr(,"class")
[1] "coa" "bga"
[1] "Data (original) range:  -0.92 0.8"
[1] "Data (scale) range:  -1.15 1.15"
[1] "Data scaled to range:  -1.15 1.15"