Normalisation "average"

[vguillemot]

Voici ce qu'il me semble correspondait à cette normalisation (extrait du manuel de BrB Array Tools) :

Normalization by reference array

This normalization method is a combination of “Median normalization of single channel data” and “Housekeeping gene normalization” in versions prior to v4.2, where a reference array needs to be selected so as to be normalized against all other arrays. The normalization is performed by computing a gene-by-gene difference between each array and the reference array, and subtracting the median difference from the log-intensities on that array, so that the gene-by-gene difference between the normalized array and the reference array is 0.

The user has the option to explicitly select one of the arrays to be the reference array, or ask BRB-ArrayTools to automatically select the “median” array as the reference array. The algorithm which BRB-ArrayTools uses to select the “median” array is as follows:

1. Let N be the number of experiments, and let i be an index of experiments running from 1 to N.
2. For each array i, the median log-intensity of the array (denoted Mi) will be computed.
3. A median M will be selected from the {M1, …, MN} values. If N is even, then the median M will be the lower of the two middle values.
4. The array whose median log-intensity Mi equals the overall median M will be chosen as the median array.

For this normalization option, the user may also be able to specify a housekeeping genelist, over which the median difference will be calculated instead of the full set of probes/probe sets.

@guegan est-ce que c'est bien cela qu'il faut implémenter dans normalisation_lumi.R ?

[guegan]

Pour la normalisation average de BeadStudio

_BeadStudio Algorithms_

Before the researcher applies an algorithm to the raw data, a reference group must be defined. A reference group can consist of one or more arrays. The default reference group is the first array or group in the list, but any array or group can be specified as the reference. The reference group is used to calculate a virtual array. A virtual array comprises the average values from all the arrays in the reference group and is used to determine normalization parameters. Once the reference group is defined for normalization, the same group is used as the reference group during differential expression analysis.

_Average Normalization_ Average normalization is used to rescale intensities across multiple arrays and chips. This algorithm is an appropriate choice for experiments that employ a large number of arrays with differences in overall intensity. For average normalization, a scaling factor, Si, is calculated by dividing the average intensity of the virtual array (μv) by the average intensity for all arrays in a group (μi). Si = μv / μi The bead-type intensities of all arrays in the experiment are normalized by Si. Ipi(norm) = IpiSi Each probe is indexed by p and the number of probes ranges from 1 to n. Each array is indexed by i and the number of arrays ranges from 1 to m.

@vguillemot JCC est en vacances jusqu'au 24. J'implémente la "méthode BeadStudio" et suivant sa réponse je modifierai en "BrB average"

[guegan]

@vguillemot Réponse de KT à mon mail : c'est bien la normalisation de BeadStudio.

[guegan]

Après test de plusieurs normalisation average sur la base de la doc BeadStudio, le résultat n'est pas concluant car on ne reproduit pas les données de JCCorvol. On attend plus d'info de leur part

[guegan]

Décision : si les utilisateurs veulent faire la normalisation average ils doivent la faire dans BeadStudio puis importer leur donner via SIMA qui transformera juste en log2. fonction lumi_normalization : bg.method="none", norm.method="none"