pareto scaling of RMA normalised values for Principle component analysis (PCA)

Question

[rajasekargutha]

how the RMA values of gene set matrix from different microarrays experiments are scale adjusted using pareto scaling which Reduce relative importance of large values, partially preservedata structure for PCA analysis.

Answer 1

[Mikael Huss]

Pareto scaling comes down to scaling each column (dividing each value in the column) by the square root of the standard deviation of the column's values.

You could adapt the code from here as appropriate:

https://github.com/cran/RFmarkerDetector/blob/master/R/scaling.R

[rajasekargutha]

its not working for me. can you suggest simple r code for doing it

[Mikael Huss]

How can I do that when I don't know what your matrix looks like? It should be something like

pareto.matrix <- apply(matrix, 2, function(x) x/sqrt(sd(x)))

[rajasekargutha]

probes GSM1122072_L1._A1_Zebrafish_.CEL GSM1122073_L10._D1_Zebrafish_.CEL GSM1122074_L11._D2_Zebrafish_.CEL GSM1122075_L12._D3_Zebrafish_.CEL GSM1122076_L13._E1_Zebrafish_.CEL GSM1122077_L14._E2_Zebrafish_.CEL GSM1122078_L15._E3_Zebrafish_.CEL GSM1122079_L2._A2_Zebrafish_.CEL GSM1122080_L3._A3_Zebrafish_.CEL GSM1122081_L4._B1_Zebrafish_.CEL GSM1122082_L5._B2_Zebrafish_.CEL AFFX-BioB-3_at 8.4878 8.7854 8.5369 8.1721 8.5262 8.2877 8.3841 8.3143 8.5436 8.4567 8.8609 AFFX-BioB-5_at 9.0681 9.3504 9.1175 8.9018 9.1054 9.0152 8.8859 9.1335 9.3054 8.9913 9.259 AFFX-BioB-M_at 8.9599 9.3359 8.8869 8.772 9.0131 8.8379 8.7329 8.9295 8.9183 9.0148 9.2427 AFFX-BioC-3_at 9.4894 9.8127 9.7298 9.5067 9.5246 9.6158 9.5922 9.506 9.5356 9.6293 9.9914 AFFX-BioC-5_at 9.5423 9.9212 9.5062 9.3627 9.5976 9.6181 9.5605 9.4079 9.4465 9.5905 9.9216 AFFX-BioDn-3_at 11.456 11.7272 11.4584 11.3387 11.4305 11.5249 11.4521 11.4418 11.469 11.4803 11.7177 AFFX-BioDn-5_at 10.6942 10.9862 10.7476 10.6058 10.703 10.6942 10.6495 10.6299 10.7293 10.7802 10.9838

[Mikael Huss]

Center by using e g

centered <- scale(matrix, center=TRUE, scale=FALSE)

(or centered <- apply(matrix, 2, function(x) x - mean(x)) as in the code snipped I linked)

Then Pareto scale using the above

pareto.matrix <- apply(centered, 2, function(x) x/sqrt(sd(x)))

[rajasekargutha]

columns are samples and rows are genes