Question: DESeq2: NA on padj column when doing bonferroni
0
5.8 years ago by
David Rengel70
European Union
David Rengel70 wrote:
Hi, I would be grateful if you could help me out on this one. Why do I get NA for certain genes on the padj column when I do this: results(dds,resultsNames(dds)[i],pAdjustMethod="bonferroni") I do not get them when I do BH Thanks for your help Best, -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]]
• 6.2k views
modified 5.8 years ago by Kristina M Fontanez220 • written 5.8 years ago by David Rengel70
0
5.8 years ago by
Kristina M Fontanez220 wrote:
Hi David- If you look at the description of results() in the manual then you will see that by default independentFiltering is turned ON. I think this is your likely culprit given that your results function call doesnt explicitly turn OFF independent filtering. pg.27: "By default, independent filtering is performed to select a set of genes which will result in the most genes with adjusted p-values less than a threshold, alpha. The adjusted p-values for the genes which do not pass the filter threshold are set to NA.  Best, Kristina ------------------------------------------------------------------ Kristina Fontanez, Postdoctoral Fellow fontanez@mit.edu<mailto:fontanez@mit.edu> Massachusetts Institute of Technology Department of Civil and Environmental Engineering 48-120E 15 Vassar Street Cambridge, MA 02139 On Jan 17, 2014, at 11:20 AM, David Rengel <david.rengel@toulouse.inra.fr<mailto:david.rengel@toulouse.inra.fr>> wrote: Hi, I would be grateful if you could help me out on this one. Why do I get NA for certain genes on the padj column when I do this: results(dds,resultsNames(dds)[i],pAdjustMethod="bonferroni") I do not get them when I do BH Thanks for your help Best, -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr<mailto:david.rengel@toulouse.inra.fr> Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]] _______________________________________________ Bioconductor mailing list Bioconductor@r-project.org https://stat.ethz.ch/mailman/listinfo/bioconductor Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor [[alternative HTML version deleted]]
Hi Kristina, That was it indeed, thanks a lot. But then again, why do I not get them, the NAs I mean, when I do BH on the same data and the same contrast? That is: results(dds,resultsNames(dds)[i],pAdjustMethod="BH") Thanks again, Best, David Le 17/01/2014 17:28, Kristina M Fontanez a écrit : > Hi David- > > If you look at the description of results() in the manual then you > will see that by default independentFiltering is turned ON. I think > this is your likely culprit given that your results function call > doesnt explicitly turn OFF independent filtering. > > pg.27: "By default, independent filtering is performed to select a set > of genes which will result in the most genes with adjusted p-values > less than a threshold, alpha. The adjusted p-values for the genes > which do not pass the filter threshold are set to NA.  > > Best, > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu <mailto:fontanez@mit.edu> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > > On Jan 17, 2014, at 11:20 AM, David Rengel > <david.rengel@toulouse.inra.fr <mailto:david.rengel@toulouse.inra.fr="">> > wrote: > >> Hi, >> >> I would be grateful if you could help me out on this one. Why do I get >> NA for certain genes on the padj column when I do this: >> >> >> results(dds,resultsNames(dds)[i],pAdjustMethod="bonferroni") >> >> >> I do not get them when I do BH >> >> >> Thanks for your help >> >> >> Best, >> >> -- >> David Rengel >> Laboratoire des Interactions Plantes Micro-organismes (LIPM) >> INRA/CNRS >> 24 Chemin de Borde Rouge >> Auzeville >> CS 52627 >> 31326 Castanet Tolosan Cedex >> >> E mail: david.rengel@toulouse.inra.fr >> <mailto:david.rengel@toulouse.inra.fr> >> Tel 33 (0)5 61 28 55 91 >> Fax 33 (0)5 61 28 50 61 >> >> http://www.toulouse.inra.fr/lipm >> >> >> [[alternative HTML version deleted]] >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor@r-project.org >> https://stat.ethz.ch/mailman/listinfo/bioconductor >> Search the archives: >> http://news.gmane.org/gmane.science.biology.informatics.conductor > -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]]
Hi David, Hi Kristina, That was it indeed, thanks a lot. But then again, why do I not get them, the NAs I mean, when I do BH on the same data and the same contrast? That is: results(dds,resultsNames(dds)[i],pAdjustMethod="BH") Thanks again, Best, David One likely possibility: Independent Filtering is selecting the set of genes that results in the most genes with adjusted p-values less than alpha 0.1 (by default ON and ON in your function call above). Any genes with adjusted p-values that do not pass filter are set to NA. If under the Bonferonni correction there are genes with adjusted p-values above 0.1 then you will have NAs in your results. However, if under BH all of your genes fall under the alpha 0.1 threshold, then you will have zero NAs in your results. Kristina ------------------------------------------------------------------ Kristina Fontanez, Postdoctoral Fellow fontanez@mit.edu<mailto:fontanez@mit.edu> Massachusetts Institute of Technology Department of Civil and Environmental Engineering 48-120E 15 Vassar Street Cambridge, MA 02139 [[alternative HTML version deleted]]
David- Im sorry but I didnt really understand your question. I dont know what you mean by Carboniferous, raw value (p-value?) or pad (p-adjusted value?). However, this section (pg 8) of the vignette may help you determine why you have NAs in the p-value column or the p-adjusted column in your results() output (see pasted below). I would spend some time reading the vignette and manual to help you answer these types of questions. Ive had to read them over several times to get the gist of things and Im still discovering new details! The results for particular genes can be set to NA, for either one of the following reasons: 1. If within a row, all samples have zero counts, this is recorded in mcols(dds)$allZero and log2fold change estimates, p-value and adjusted p-value will all be set to NA. 2. If a row contains a sample with an extreme count then the p-value and adjusted p-value are set to NA. These outlier counts are detected by Cooks distance. Customization of this outlier filteringis described in Section 3.3, along with a method for replacing outlier counts and refitting. 3. If a row is filtered by automatic independent filtering, then only the adjusted p-value is set to NA. Description and customization of independent filtering is decribed in Section 3.6. Kristina ------------------------------------------------------------------ Kristina Fontanez, Postdoctoral Fellow fontanez@mit.edu<mailto:fontanez@mit.edu> Massachusetts Institute of Technology Department of Civil and Environmental Engineering 48-120E 15 Vassar Street Cambridge, MA 02139 On Jan 17, 2014, at 3:11 PM, David Rengel <david.rengel@toulouse.inra.fr<mailto:david.rengel@toulouse.inra.fr>> wrote: Thanks Kristina and Jose Manuel for your helpful replies. What I do not fully understand about the filtering is the following: After Carboniferous, there are ~3K Ans with non Ans values on the raw Value column, and that is out of ~2OK total genes, the range of those raw Values go from 2.95E-05 to 0.99. And there are other genes with similar Values that do not give NA on pad. And, of course, I do have genes that show 1 on the padj column. It is a bit puzzling to me. I guess it depends on their level of expression? Thanks! Best, David Le 17/01/2014 17:57, Kristina M Fontanez a écrit : Hi David, Hi Kristina, That was it indeed, thanks a lot. But then again, why do I not get them, the NAs I mean, when I do BH on the same data and the same contrast? That is: results(dds,resultsNames(dds)[i],pAdjustMethod="BH") Thanks again, Best, David One likely possibility: Independent Filtering is selecting the set of genes that results in the most genes with adjusted p-values less than alpha 0.1 (by default ON and ON in your function call above). Any genes with adjusted p-values that do not pass filter are set to NA. If under the Bonferonni correction there are genes with adjusted p-values above 0.1 then you will have NAs in your results. However, if under BH all of your genes fall under the alpha 0.1 threshold, then you will have zero NAs in your results. Kristina ------------------------------------------------------------------ Kristina Fontanez, Postdoctoral Fellow fontanez@mit.edu<mailto:fontanez@mit.edu> Massachusetts Institute of Technology Department of Civil and Environmental Engineering 48-120E 15 Vassar Street Cambridge, MA 02139 -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr<mailto:david.rengel@toulouse.inra.fr> Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]] ADD REPLYlink written 5.8 years ago by Kristina M Fontanez220 hi David, Yes, I think most people have touched on the answer here. And please take a look at the vignette, especially the Theory section on independent filtering. We've spent some time in writing the vignette trying to break down what is going on. Your question was, why do some genes have NA for the adjusted p-value with Bonferroni correction but not for BH correction. The answer is that the genefilter package which we use to perform independent filtering uses a vector called 'filter' to limit the number of tests which will have p-values adjusted for multiple testing. The decision is made by picking a value which maximizes #(adjusted p-values < alpha). The adjustment methods are different and produce different values. Bonferroni is more conservative for example. So the filtering, which depends as a function on the underlying method for p-value adjustment, will behave differently for different adjustment methods. Mike On Fri, Jan 17, 2014 at 3:22 PM, Kristina M Fontanez <fontanez@mit.edu>wrote: > David- > > Iâm sorry but I didnât really understand your question. I donât know what > you mean by Carboniferous, raw value (p-value?) or pad (p-adjusted value?). > > However, this section (pg 8) of the vignette may help you determine why > you have NAs in the p-value column or the p-adjusted column in your > results() output (see pasted below). I would spend some time reading the > vignette and manual to help you answer these types of questions. Iâve had > to read them over several times to get the gist of things and Iâm still > discovering new details! > > > The results for particular genes can be set to NA, for either one of the > following reasons: > 1. If within a row, all samples have zero counts, this is recorded in > mcols(dds)$allZero and log2fold change estimates, p-value and adjusted > p-value will all be set to NA. > > 2. If a row contains a sample with an extreme count then the p-value and > adjusted p-value are set to NA. These outlier counts are detected by Cookâs > distance. Customization of this outlier filteringis described in Section > 3.3, along with a method for replacing outlier counts and refitting. > > 3. If a row is filtered by automatic independent filtering, then only the > adjusted p-value is set to NA. Description and customization of independent > filtering is decribed in Section 3.6. > > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu<mailto:fontanez@mit.edu> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > > On Jan 17, 2014, at 3:11 PM, David Rengel <david.rengel@toulouse.inra.fr> <mailto:david.rengel@toulouse.inra.fr>> wrote: > > Thanks Kristina and Jose Manuel for your helpful replies. > What I do not fully understand about the filtering is the following: After > Carboniferous, there are ~3K Ans with non Ans values on the raw Value > column, and that is out of ~2OK total genes, the range of those raw Values > go from 2.95E-05 to 0.99. And there are other genes with similar Values > that do not give NA on pad. And, of course, I do have genes that show 1 on > the padj column. > It is a bit puzzling to me. I guess it depends on their level of > expression? > > Thanks! > > Best, > > David > > Le 17/01/2014 17:57, Kristina M Fontanez a Ã©crit : > > Hi David, > > Hi Kristina, > > That was it indeed, thanks a lot. But then again, why do I not get them, > the NAs I mean, when I do BH on the same data and the same contrast? That > is: > results(dds,resultsNames(dds)[i],pAdjustMethod="BH") > Thanks again, > Best, > David > > > One likely possibility: > Independent Filtering is selecting the set of genes that results in the > most genes with adjusted p-values less than alpha 0.1 (by default ON and ON > in your function call above). Any genes with adjusted p-values that do not > pass filter are set to NA. If under the Bonferonni correction there are > genes with adjusted p-values above 0.1 then you will have NAs in your > results. However, if under BH all of your genes fall under the alpha 0.1 > threshold, then you will have zero NAs in your results. > > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu<mailto:fontanez@mit.edu> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > -- > David Rengel > Laboratoire des Interactions Plantes Micro-organismes (LIPM) > INRA/CNRS > 24 Chemin de Borde Rouge > Auzeville > CS 52627 > 31326 Castanet Tolosan Cedex > > E mail: david.rengel@toulouse.inra.fr<mailto:david.rengel@toulouse.inra.fr> > > Tel 33 (0)5 61 28 55 91 > Fax 33 (0)5 61 28 50 61 > > http://www.toulouse.inra.fr/lipm > > > > > > [[alternative HTML version deleted]] > > > _______________________________________________ > Bioconductor mailing list > Bioconductor@r-project.org > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor > [[alternative HTML version deleted]]
Thank you Mike for your helpful and clear reply. I'll have a closer look at the vignette as you suggested. Best wishes, David Le 17/01/2014 22:04, Michael Love a Ã©crit : > hi David, > > Yes, I think most people have touched on the answer here. And please > take a look at the vignette, especially the Theory section on > independent filtering. We've spent some time in writing the vignette > trying to break down what is going on. > > Your question was, why do some genes have NA for the adjusted p-value > with Bonferroni correction but not for BH correction. The answer is > that the genefilter package which we use to perform independent > filtering uses a vector called 'filter' to limit the number of tests > which will have p-values adjusted for multiple testing. The decision > is made by picking a value which maximizes #(adjusted p-values < > alpha). The adjustment methods are different and produce different > values. Bonferroni is more conservative for example. So the filtering, > which depends as a function on the underlying method for p-value > adjustment, will behave differently for different adjustment methods. > > Mike > > > On Fri, Jan 17, 2014 at 3:22 PM, Kristina M Fontanez <fontanez@mit.edu> <mailto:fontanez@mit.edu>> wrote: > > David- > > Iâm sorry but I didnât really understand your question. I donât > know what you mean by Carboniferous, raw value (p-value?) or pad > (p-adjusted value?). > > However, this section (pg 8) of the vignette may help you > determine why you have NAs in the p-value column or the p-adjusted > column in your results() output (see pasted below). I would spend > some time reading the vignette and manual to help you answer these > types of questions. Iâve had to read them over several times to > get the gist of things and Iâm still discovering new details! > > > The results for particular genes can be set to NA, for either one > of the following reasons: > 1. If within a row, all samples have zero counts, this is recorded > in mcols(dds)$allZero and log2fold change estimates, p-value and > adjusted p-value will all be set to NA. > > 2. If a row contains a sample with an extreme count then the > p-value and adjusted p-value are set to NA. These outlier counts > are detected by Cookâs distance. Customization of this outlier > filteringis described in Section 3.3, along with a method for > replacing outlier counts and refitting. > > 3. If a row is filtered by automatic independent filtering, then > only the adjusted p-value is set to NA. Description and > customization of independent filtering is decribed in Section 3.6. > > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu <mailto:fontanez@mit.edu><mailto:fontanez@mit.edu> <mailto:fontanez@mit.edu>> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > > On Jan 17, 2014, at 3:11 PM, David Rengel > <david.rengel@toulouse.inra.fr> <mailto:david.rengel@toulouse.inra.fr><mailto:david.rengel@toulo use.inra.fr=""> <mailto:david.rengel@toulouse.inra.fr>>> wrote: > > Thanks Kristina and Jose Manuel for your helpful replies. > What I do not fully understand about the filtering is the > following: After Carboniferous, there are ~3K Ans with non Ans > values on the raw Value column, and that is out of ~2OK total > genes, the range of those raw Values go from 2.95E-05 to 0.99. And > there are other genes with similar Values that do not give NA on > pad. And, of course, I do have genes that show 1 on the padj column. > It is a bit puzzling to me. I guess it depends on their level of > expression? > > Thanks! > > Best, > > David > > Le 17/01/2014 17:57, Kristina M Fontanez a Ã©crit : > > Hi David, > > Hi Kristina, > > That was it indeed, thanks a lot. But then again, why do I not get > them, the NAs I mean, when I do BH on the same data and the same > contrast? That is: > results(dds,resultsNames(dds)[i],pAdjustMethod="BH") > Thanks again, > Best, > David > > > One likely possibility: > Independent Filtering is selecting the set of genes that results > in the most genes with adjusted p-values less than alpha 0.1 (by > default ON and ON in your function call above). Any genes with > adjusted p-values that do not pass filter are set to NA. If under > the Bonferonni correction there are genes with adjusted p-values > above 0.1 then you will have NAs in your results. However, if > under BH all of your genes fall under the alpha 0.1 threshold, > then you will have zero NAs in your results. > > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu <mailto:fontanez@mit.edu><mailto:fontanez@mit.edu> <mailto:fontanez@mit.edu>> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > -- > David Rengel > Laboratoire des Interactions Plantes Micro-organismes (LIPM) > INRA/CNRS > 24 Chemin de Borde Rouge > Auzeville > CS 52627 > 31326 Castanet Tolosan Cedex > > E mail: david.rengel@toulouse.inra.fr > <mailto:david.rengel@toulouse.inra.fr><mailto:david.rengel@toulo use.inra.fr=""> <mailto:david.rengel@toulouse.inra.fr>> > Tel 33 (0)5 61 28 55 91 > Fax 33 (0)5 61 28 50 61 > > http://www.toulouse.inra.fr/lipm > > > > > > [[alternative HTML version deleted]] > > > _______________________________________________ > Bioconductor mailing list > Bioconductor@r-project.org <mailto:bioconductor@r-project.org> > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor > > -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]] ADD REPLYlink written 5.8 years ago by David Rengel70 Thanks again Kristina. I'll have a close look at it. PS: I do not know where on Earth the "Carboniferous" appeared from. What I meant was "Bonferroni". Funny. Le 17/01/2014 21:22, Kristina M Fontanez a écrit : > David- > > Im sorry but I didnt really understand your question. I dont know > what you mean by Carboniferous, raw value (p-value?) or pad > (p-adjusted value?). > > However, this section (pg 8) of the vignette may help you determine > why you have NAs in the p-value column or the p-adjusted column in > your results() output (see pasted below). I would spend some time > reading the vignette and manual to help you answer these types of > questions. Ive had to read them over several times to get the gist of > things and Im still discovering new details! > > The results for particular genes can be set to NA, for either one of > the following reasons: > 1. If within a row, all samples have zero counts, this is recorded in > mcols(dds)$allZero and log2fold change estimates, p-value and adjusted > p-value will all be set to NA. > > 2. If a row contains a sample with an extreme count then the p-value > and adjusted p-value are set to NA. These outlier counts are detected > by Cooks distance. Customization of this outlier filteringis > described in Section 3.3, along with a method for replacing outlier > counts and refitting. > > 3. If a row is filtered by automatic independent filtering, then only > the adjusted p-value is set to NA. Description and customization of > independent filtering is decribed in Section 3.6. > > > Kristina > ------------------------------------------------------------------ > Kristina Fontanez, Postdoctoral Fellow > fontanez@mit.edu <mailto:fontanez@mit.edu> > Massachusetts Institute of Technology > Department of Civil and Environmental Engineering > 48-120E > 15 Vassar Street > Cambridge, MA 02139 > > > > On Jan 17, 2014, at 3:11 PM, David Rengel > <david.rengel@toulouse.inra.fr <mailto:david.rengel@toulouse.inra.fr="">> > wrote: > >> Thanks Kristina and Jose Manuel for your helpful replies. >> What I do not fully understand about the filtering is the following: >> After Carboniferous, there are ~3K Ans with non Ans values on the >> raw Value column, and that is out of ~2OK total genes, the range of >> those raw Values go from 2.95E-05 to 0.99. And there are other genes >> with similar Values that do not give NA on pad. And, of course, I do >> have genes that show 1 on the padj column. >> It is a bit puzzling to me. I guess it depends on their level of >> expression? >> >> Thanks! >> >> Best, >> >> David >> >> Le 17/01/2014 17:57, Kristina M Fontanez a écrit : >>> >>> Hi David, >>> >>>> Hi Kristina, >>>> >>>> That was it indeed, thanks a lot. But then again, why do I not get >>>> them, the NAs I mean, when I do BH on the same data and the same >>>> contrast? That is: >>>> results(dds,resultsNames(dds)[i],pAdjustMethod="BH") >>>> Thanks again, >>>> Best, >>>> David >>> >>> >>> One likely possibility: >>> Independent Filtering is selecting the set of genes that results in >>> the most genes with adjusted p-values less than alpha 0.1 (by >>> default ON and ON in your function call above). Any genes with >>> adjusted p-values that do not pass filter are set to NA. If under >>> the Bonferonni correction there are genes with adjusted p-values >>> above 0.1 then you will have NAs in your results. However, if under >>> BH all of your genes fall under the alpha 0.1 threshold, then you >>> will have zero NAs in your results. >>> >>> Kristina >>> ------------------------------------------------------------------ >>> Kristina Fontanez, Postdoctoral Fellow >>> fontanez@mit.edu <mailto:fontanez@mit.edu> >>> Massachusetts Institute of Technology >>> Department of Civil and Environmental Engineering >>> 48-120E >>> 15 Vassar Street >>> Cambridge, MA 02139 >> >> -- >> David Rengel >> Laboratoire des Interactions Plantes Micro-organismes (LIPM) >> INRA/CNRS >> 24 Chemin de Borde Rouge >> Auzeville >> CS 52627 >> 31326 Castanet Tolosan Cedex >> >> E mail:david.rengel@toulouse.inra.fr >> Tel 33 (0)5 61 28 55 91 >> Fax 33 (0)5 61 28 50 61 >> >> http://www.toulouse.inra.fr/lipm >> >> > -- David Rengel Laboratoire des Interactions Plantes Micro-organismes (LIPM) INRA/CNRS 24 Chemin de Borde Rouge Auzeville CS 52627 31326 Castanet Tolosan Cedex E mail: david.rengel@toulouse.inra.fr Tel 33 (0)5 61 28 55 91 Fax 33 (0)5 61 28 50 61 http://www.toulouse.inra.fr/lipm [[alternative HTML version deleted]]