[BioC] (no subject)

Gordon K Smyth smyth at wehi.EDU.AU
Fri Aug 10 01:52:46 CEST 2012

> Date: Wed, 8 Aug 2012 19:55:11 +0200
> From: Amos Kirilovsky <amos.kirilovsky at gmail.com>
> To: Matthew Ritchie <mritchie at wehi.edu.au>
> Cc: bioconductor at r-project.org
> Subject: Re: [BioC] Limma, arrayWeights and fold change
> Dear Matt,
> Thank you very much for your answer. I'm glad the function works with
> biological replicates. Concerning the other points, you said that one
> group is more consistent than the other one. When this appears, can
> arrayWeights introduce a bias (more than if not using the function)? If
> yes, what can I do?

No, there is no bias.

> Moreover, if I use logFC to make heatmap or anything else, should I use
> logFC as estimated when using arrayWeights or should I use the regular fold
> change?

Naturally you should use the logFC using arrayWeights, because they take 
into account the differential quality and the arrays and are therefore 
more accurate.

> My quality controls and arrayWeights  showed that some arrays have poor
> quality. ArrayWeights allow me to keep them when detecting differentially
> expressed genes. Is there a way to take into account the quality when doing
> other tests as clustering, correlations... ?

Almost all limma functions automatically use the estimated weights. 
Functions in other packages will generally not do so.


> Best,
> Amos
> 2012/8/4 Matthew Ritchie <mritchie at wehi.edu.au>:
>> Dear Amos,
>>> I'm using the limma package and the arrayWeights function to make some
>>> transcriptomic analysis: One group of samples versus an another group.
>>> arrayWeights allow me to get very interesting results but in
>>> documentation it says you need replicates. Are they technical or/ and
>>> biological replicates? I couldn't find clearly the information.
>> Ideally biological replicates.  Technical replicates are handled by other
>> functions in limma, such as duplicateCorrelation().
>>> My second question is: can arrayWeights introduce some bias? for
>>> example in one of my analysis with 2 groups and 5 samples by group,
>>> the weights found by arrayWeights are:
>>> for the 1st group : 2.91 0.99 0.43 2.18 0.89
>>> for the 2nd group : 1.40 1.24 0.29 1 .40 0.60
>>> It seems that the weights are in general higher for the first group.
>>> Is that a problem?
>> This would indicate that the first group of arrays are on average more
>> consistent than the second group of arrays.
>>> Next in limma, when Weights calculated by arrayWeights are implemented
>>> to the the lmFit function how are calculated the log2 fold change
>>> (coefficients ) ?
>>> I tried to calculate them by myself applying the weights to the
>>> samples but I couldn't get the same results.
>>> I observed that the average expression of each probe with or without
>>> applying the weights is equal with the function topTable. Is that
>>> normal?
>> The weights are used in gene-wise weighted least squares regression to
>> estimate the coefficients in the linear model.  So they will affect the
>> logFC values.  They are not used in the calculation of average expression,
>> so these value won't be affected by the use of weights.
>> I hope this helps.  Best wishes,
>> Matt

The information in this email is confidential and intend...{{dropped:4}}

More information about the Bioconductor mailing list