Question

Selection of appropriate gene filtering method for discovering statistically significant genes for Affymetrix arrays

0

Entering edit mode

svlachavas ▴ 830

@svlachavas-7225

Last seen 6 months ago

Germany/Heidelberg/German Cancer Resear…

I have been preprocessing in R a big data set in order to identify possible differentiated genes for two conditions. My main problem-question is, as i searched in literature, some people using filtering based on present/absent calls to remove probes which are totally(or in a significant persent) absent in their arrays. On the other hand, an important number of methodologies after normalization and quality control, use various implementations of non-specific filtering(for instance based on variance) prior to different statistical tests. My Affymetrix platform is HG-U133 plus2.0 array.

bioconductor differential gene expression independent filtering • 1.9k views

ADD COMMENT • link 9.3 years ago svlachavas ▴ 830

score 1 · Answer 1 · 2015-01-10

1

Entering edit mode

Gordon Smyth 50k

@gordon-smyth

Last seen 5 hours ago

WEHI, Melbourne, Australia

How are you intending to assess differential expression? If you intending to use limma, then you may filter on present/absence calls or an average log-expression, but you must not filter on variance.

All filtering methods that are unsupervised in terms of the sample allocations are considered to be non-specific for filtering purposes, including present/absence calls and average expression.

ADD COMMENT • link 9.3 years ago Gordon Smyth 50k

0

Entering edit mode

For one first approach i want to compare the 34 samples i have into two conditions to compare from the phenoData. I have tried to perform limma based on variance, but i have read from other threads, as also from other papers that is not recommended to combine variance filtering with limma. On the other hand, if i choose to filter on absent/present calls, in Affymetrix i have roughly two main options: mas5calls & the panp package which can be used for my specific platform. I have used the commands to generate present absent calls, but my main question is if i have also to filter out marginal calls or leave them in my ExpressionSet ? Also other methods i could implement based on this specific big dataset, could be multiple test procedure with unequal variance and fdr correction or the SAM test ? i could paste here a small sample from my script to give me an opinion about filtering out based in absent/present calls.Thank you again for your consideration on this matter !!

Best regards

ADD REPLY • link 9.3 years ago svlachavas ▴ 830

0

Entering edit mode

You seem to be asking the same questions on two different threads, see Non-specific filtering methodogies for ExpressionSet in R/Bioconductor

ADD REPLY • link 9.3 years ago Gordon Smyth 50k

score 0 · Answer 2 · 2015-01-12

Dear Gordon,

Please excuse me because i didnt intend to ask for the same question and maybe i was misunderstood !! My basic question(problem) is whether i should use present/absent calls with Limma !! And secondly, if i use some filtering on present/absent calls, in your opinion if i want to remove uninformative probe-sets with Absent calls, what arbitary cut-off should i use ? For instance, i have 34 CEL files/samples in my dataset. So should i filter out these probesets with more of 80% absent calls ? Moreover, after my filtering, is useful to use some algorithm(like kNN) for missing value imputation ?

Best regards