AEsetnorm quality metrics report

Summary

Array #	Array Name	Scan Dates	MA plots	Boxplots	Heatmap
1	459F.CEL	04/22/04 10:52:15
2	459D.CEL	04/22/04 09:35:41
3	459B.CEL	04/22/04 09:23:34
4	515J.CEL	08/13/04 13:32:20
5	515G.CEL	08/13/04 12:12:39
6	459A.CEL	04/22/04 10:25:46
7	459C.CEL	04/22/04 09:11:38
8	515D.CEL	08/13/04 10:06:35
9	515I.CEL	08/13/04 13:56:55
10	515E.CEL	08/13/04 12:00:12
11	515K.CEL	08/13/04 13:44:35
12	515H.CEL	08/13/04 12:25:05
13	515C.CEL	08/13/04 10:19:00
14	515F.CEL	08/13/04 11:47:47
15	515B.CEL	08/13/04 10:32:33
16	515L.CEL	08/13/04 14:09:24

*outlier array

Index

PLEASE NOTE:
All figures below are links to PDF files: these contain images for every array in the report. The PDF files may be several pages long, this HTML report presents only the first page.

Section 1: Individual array quality
Section 2: Array intensity distributions
Section 3: Between array comparison
Section 4: Variance mean dependence

Section 1: Individual array quality

Figure 1: MA plots

Figure 1 shows the MA plot for each array. M and A are defined as :
M = log₂(I₁) - log₂(I₂)
A = 1/2 (log₂(I₁)+log₂(I₂)),
where I₁ is the intensity of the array studied and I₂ is the intensity of a "pseudo"-array, which have the median values of all the arrays. Typically, we expect the mass of the distribution in an MA plot to be concentrated along the M = 0 axis, and there should be no trend in the mean of M as a function of A. A trend in the lower range of A usually indicates that the arrays have different background intensities, this may be addressed by background correction. A trend in the upper range of A usually indicates saturation of the measurements, in mild cases, this may be addressed by non-linear normalisation (e.g. quantile normalisation).

Section 2: Array intensity distributions

Figure 2: Boxplots

Figure 2 presents boxplots of the log₂(Intensities). Each box corresponds to one array. It gives a simple summary of the distribution of probe intensities across all arrays. Typically, one expects the boxes to have similar size (IQR) and y position (median). If the distribution of an individual array is very different from the others, this may indicate an experimental problem. After normalisation, the distributions should be similar.

Figure 3: Density plots

Figure 3 shows density estimates (smoothed histograms) of the data. Typically, the distributions of the arrays should have similar shapes and ranges. Arrays whose distributions are very different from the others should be considered for possible problems. On raw data, a bimodal distribution can be indicative of an array containing a spatial artefact and an array shifted to the right of an array with abnormal higher background intensities.

Section 3: Between array comparison

Figure 4: Heatmap representation of the distance between arrays

Figure 4 shows a false colour heatmap of between arrays distances, computed as the mean absolute difference (L₁-distance) of the vector of M-values for each pair of arrays on every probes without any filtering. The colour scale is chosen to cover the range of L₁-distances encountered in the dataset. Arrays for which the sum of the distances to the others is much different from the others, are detected as outlier arrays. The dendrogram on this plot also can serve to check if, the arrays cluster accordingly to a biological meaning.
d_xy = mean|M_xi-M_yi|
Here, M_xi is the M-value of the i-th probe on the x-th array, without preprocessing. Consider the following decomposition of M_xi: M_xi = z_i + β_xi + ε_xi, where z_i is the probe effect for probe i (the same across all arrays), ε_xi are i.i.d. random variables with mean zero and β_xi is such that for any array x, the majority of values β_xi are negligibly small (i. e. close to zero). β_xi represents differential expression effects. In this model, all values d_xy are (in expectation) the same, namely 2 times the standard deviation of ε_xi.

Figure 5: Principal Component Analysis

Figure 5 represents a biplot for the first two principal components from the dataset. The colours correspond to the group of interest given. We expect the arrays to cluster accordingly to a relevant experimental factor. The principal components transformation of a data matrix re-expresses the features using linear combination of the original variables. The first principal component is the linear combination chosen to possess maximal variance, the second is the linear combination orthogonal to the first possessing maximal variance among all orthogonal combination.

Section 4: Variance mean dependence

Figure 6: Standard deviation versus rank of the mean

For each feature, Figure 6 shows the standard deviation of the intensities across arrays on the y-axis versus the rank of their mean on the x-axis. The red dots, connected by lines, show the running median of the standard deviation. After normalisation and transformation to a logarithm(-like) scale, one typically expects the red line to be approximately horizontal, that is, show no substantial trend. In some cases, a hump on the right hand of the x-axis can be observed and is symptomatic of a saturation of the intensities.

This report has been created with arrayQualityMetrics 2.6.0 under R version 2.11.0 (2010-04-22)

(Page generated on Thu Apr 22 18:14:12 2010 by hwriter )