www.bcb.lon.ac.uk

Quality Control Analysis Sample Report

Summary

Data Pre-Processing

Quality Control Plots:
   Boxplots
   Reproducibility between replicates
   PMvsMM Histograms
   RNA Degradation
   MvA Plots
   Control Probeset Analysis
   Correlation Plots
Appendix

Report Summary

Data Pre-Processing

Pre-processing of Affymetrix data involves three steps: background adjustment, normalisation and summarisation. Normalisation adjusts the data to make the measurements from different arrays comparable. Summarisation combines the multiple probe intensities for each probeset to produce a single expression value for each probe set. We use the following two methods:

1. MAS5.0 corrects for hybridisation to the mismatch (MM) probes for that particular probeset. The output is an expression value which we convert to log2 prior to further analysis. We do not filter on Absent, Present, Marginal calls. We then perform a between-chip loess normalisation on all GeneChips included in the experiment. The intensity values (log2) for the MAS5-normalised data can be found in file MAS5_loess_log2.tsv, which can be opened in Excel (.tsv is tab-separated value).


2. GCRMA is an alternative method which calculates a background adjustment step that ignores the MM intensities (RMA; Robust Multi-Array Analysis), which also incorporates sequence information from the probes (GC). This method performs within-chip and between-chip normalisations in a single step. The intensity values (log2) for the GCRMA-normalised data can be found in file GCRMA.tsv, which can be opened in Excel.

QC Plots

Boxplots

Pre-normalised Data

Normalised Data

Reproducibility between replicates (PM data only)

Histograms of PM vs MM

RNA Degradation

chip slope p-value t0_1 7.09 3.05e-14 t0_2 6.22 4.74e-13 t0_3 6.77 3.66e-13 t1_A1 5.62 7.2e-11 t1_A2 5.7 9.7e-11 t1_A3 6.3 1.24e-14 t2_A1 5.39 3.93e-13 t2_A2 5.85 1.32e-13 t2_A3 6.07 2.4e-13

MvA Plots

Control Probeset Analysis

Correlation Plots

Appendix

The data was analysed using Bioconductor v1.5 and R version 2.1.0

Bioconductor: open software development for computational biology and bioinformatics

Wu et al., GCRMA normalisation

Miller et al. Simpleaffy package

TIGR MeV v3.1

Affect of RNA Amplification on RNA Degradation

Higher degradation is to be expected when an extra round of RNA amplification has been used, so reference chips must be chosen carefully.

To illustrate this, the RNA degradation plot below was made using published data from several mouse M403_2 chips. These datasets were downloaded from the Gene Expression Omnibus, GEO. The comparison used data obtained from eight chips hybridised with Universal Mouse Reference RNA Samples from Stratagene.

n

OneRA: One round of amplification
TwoRA: Two rounds of amplification
RS   : Ribo-SPIA amplification

Samples	GSM24056-GSM24058; GSM24060-GSM24061; GSM24062-GSM24064
Slope	1.07, 1.03, 1.05, 5.35, 5.24, 4.51, 4.58, 4.48
Pvalue	3.98e-04, 7.90e-04, 5.86e-04, 1.02e-12, 4.90e-13, 1.03e-10, 9.54e-11, 8.43e-11

MvA Plots

Type of Samples and Comments	MvA Plots
GCRMA normalised data from two ‘good’ chips. Notice the almost flat Loess line (red) and the near symmetric nature of the plot about the A axis.
MAS5.0 and Loess normalised data from two good chips. Once again the plot is almost symmetric about the A axis, although notice the different shape to the plot, with larger peaks around A=4. This is characteristic of MAS5.0 normalised data and not a QC issue.
MAS5.0 normalisation on an extremely poor set of 4 chips. Notice that although there is still some of the characteristic MAS5.0 shape to the plots, the loess lines in red are not particularly straight. None of the plots is remotely symmetric about the A axis: each of the plots shows a distinct skew for higher A values.

SimpleAffy QC

A number of additional QC assessments can be made on the MAS5.0 normalised data using the BioConductor library Simpleaffy. These are as follows:

1. Average background
   This should be similar across all chips. Differences could be due to unequal amounts of RNA in the hybridisation step or due to one of the hybridisations incorporating more label, thus producing a ‘brighter’ chip. Values are given next to each chip, underneath the %Present call.


2. Scale factor
   MAS5.0 scales the intensity for each sample so that each array has the same mean; the amount of scaling required is given by the scale factor. A broad range of scale factors across the arrays would indicate differing amounts or quality of RNA. Affymetrix recommend that scale factors should lie within 3-fold of each other. The blue stripe represents this range of acceptable scale factors. Scale factors are plotted as dots on the end a line from the centre of the image. A line to the left indicates down-scaling; a line to right indicates up-scaling. Were any of the scale factors to fall outside the strip, they would all be coloured red.


3. Number of genes called ‘Present’
   By generating Present/Marginal/Absent calls, the fraction of genes called ‘Present’ (‰Present) can be calculated for each array. As before, differing amounts or quality of RNA can lead to a broad range of %Present values across the arrays. Variation between different tissue types or samples may be expected. However, the %Present values of replicates ought to be similar. Values are given to the right of each chip name, on the left hand side of the plot.


4. 3’ to 5’ ratios
   3’ to 5’ ratios provide a measure of the quality of the RNA used. Rather than the previous RNA degradation graphs which considered all probes across an array, Simpleaffy looks specifically at Β-actin and GAPDH, two relatively long genes. The reasoning is similar: by comparing the intensity values from the 3’ probeset to the 5’ or mid-point probesets a 3’:5’ ratio can be calculated. A high ratio indicates significant RNA degradation or a poor in vitro transcription (IVT) step. GAPDH 3’:5’ ratios are plotted as circles – ratios > 1 are coloured red (unacceptable). Β-actin 3’:5’ ratios are plotted as triangles. Being longer than GAPDH, unacceptable ratios are > 3 and also coloured red. Acceptable ratios for both are coloured blue.


5. Spike-in probesets
   Additional labelled cRNAs are added to validate the hybridisation step. BioB is added at a concentration of 1.5pM. This amounts to roughly 3 transcripts per cell which is the lower limit of detection for the Affymetrix system. Ideally BioB should be called ‘Present’ on every array, but a lower bound of 70% Present across all arrays is considered acceptable. This information is not plotted but can be obtained using the R command outlined above.

An example SimpleAffy QC plot is shown below. More information on these QC metrics can be found in the SimpleAffy website from which this plot was taken.