Talk:Community action for a commercial alternative to microarray analyses

From DictyWiki

As mentioned in the email message sent today, I welcome every contribution to the debate:

on 6/12/07 19:45, Gad shaulsky at gadi@bcm.edu wrote:

Dear Thierry,

I think there are a few issues that you and others should be aware of regarding Dicty microarrays. Last year, my lab spent over 8 months testing two commercial oligo array platforms. In one case we had an array of 10,000 oligos designed and printed by Agilent. The deal with Agilent was great - they printed a set of arrays and sent us half of the set, and we sent them Dicty RNA . They used our RNA on their chip with their tried-and-true methods and we used it on our cDNA array and on their oligo chip side-by-side. They only charged us 50% of the cost of the arrays and did all the analysis on their part, but unfortunately, the results were terrible. In both laboratories the oligo chip gave beautiful strong signals and great signal-to-noise ratios but there was almost no difference between elements that hybridize with abundant mRNA and elements that hybridize with rare mRNA. There was also very little difference between different developmental time points. We think this was due to high levels of cross hybridization or some non-specific hybridization. At the same time, our regular cDNA chips gave good results that correlated well with what we know about developmental genes etc. from our own microarray experience and from published Northern blots and RT-PCR data. We then tried oligo arrays that were printed by another company, Ocimum. The results were essentially as bad as the ones we got from the Agilent chip.

Bill Loomis' lab have made their own collection of 1,200 oligonucleotides and printed a chip in their facility at UCSD. They got very good results. They sent us some of their oligos and we printed them in our facility. The results in our hands were almost as good as those obtained by Bill's lab.

Last year we also looked into working with Affymetrix chips but found that the cost would be prohibitive. Combining the cost of the chip and the cost of processing in our microarray core at Baylor, we would have to pay about $1,000 per chip in addition to labor and materials for preparing the RNA samples in our lab and subsequent analysis. This would add up to about $15,000 per experiment for the most simple experimental design (see below) and about $250,000 for a typical experiment. The latter is the equivalent of a 1-year budget on an NIH R01 grant, which would not be practical.

In the most simple design of a microarray experiment, one would compare an experimental sample (e.g. Wild type cells treated with some drug) to a control (e.g. Wild type cells incubated under the same conditions without the drug). In our experience, you need 3 biological replications (independent repeats of the experiment) and 2 technical replications (two chips from each RNA sample) in order to obtain significant and reproducible results. This adds up to 6 chips from each of the two samples (treated vs. untreated), or $12,000 in chips and microarray core services alone. Realistically, you should plan on about $15,000 to include other materials, labor and a reasonable failure rate.

The above is true for the most simple experiment, but that experiment is almost never done. In reality, you would probably want to test a few conditions, such as 3-4 different drug concentrations and maybe 2-3 incubation times. If you are doing a developmental time course, you would probably want to examine a few time points as well. We usually prepare RNA from 13 time points (24 hours, every 2 hours) for a detailed analysis, or 3-7 time points for more general analyses. We repeat each experiment three times and run each sample on 2-3 chips.

We calculated the cost of Affy chips, custom oligo chips (Agilent and Ocimum) and in-house oligo chips (ordering the oligos from Operon) and found that none of them would be affordable to us. Regarding the debate about high throughput sequencing, I think that it would be even less practical. As of today, running one sequencing reaction (i.e. One experimental condition) at the Human Genome Sequencing Center here at BCM costs $7,000 on the 454 platform and $5,000 on the Solexa platform. The sequence yields are 120 Mb and 800 Mb, respectively, so you would probably get enough coverage to obviate the need for technical replication, but you would still need to test replicates to get an estimate of biological variation. The most simple experiment would therefore cost $42,000 on the 454 platform and $30,000 on the Solexa platform (RNA from treated and untreated cells in 3 biological replications each). Based on these calculations, my lab decided to continue to use cDNA arrays until the price of oligo chips comes down significantly or until a better technology appears.

I also want people to thoroughly consider the issue of data analysis. In our experience, it is hard to get good analysis from a core facility or from a company. Cluster analysis and differential expression analysis, which are what most facilities can provide, are usually not sufficient and it is very frustrating to get huge amounts of data that you cannot interpret. We do all of our analysis in-house and we have collaborations with experts in statistics and in data mining that help us a lot. It would be important to establish such collaborations with local experts before you design the experiments.

The fact that we failed to utilize an oligonucleotide platform should not discourage anyone from testing it again or from trying new methods, but I hope that you will design your experiments carefully and run a few pilot experiments before committing a lot of money to any method. My lab will be more than happy to help anyone in designing new experimental platforms, providing protocols, analyzing data, etc.

Bill Loomis and I discussed these issues and we are in agreement. Bill would be happy to share his experience and provide advice too.

Sincerely,

Gadi

on 6/12/07 18:37, Jeffrey Williams at j.g.williams@dundee.ac.uk wrote:

Dear all,

I am delighted to find that action is being taken to ensure that we will have arrays for Dicty and I promise my full support. I can't comment on the relative merits of the various system because I only have experience with our own arrays made from the Japanese cDNAs and, more indirectly, with the Sanger arrays. I guess the only way forward is to cost and compare the various systems and that will allow you to get some firm commitments from people like me. At present the range of costs quoted is very wide and per slide cost has a major effect on the scale of one's array ambitions.

Jeff

on 5/12/07 15:08, Gernot Gloeckner at gernot@fli-leibniz.de wrote:

Dear all,

we have already done some transcription profiling using 454 sequencing. Yet we did use D. fasciculatum, a group one species with an A/T content of only 66 %. In our hands this works quite well, the quantification is very straight forward. The problem here is the costs, you need at least one 454 run (~10.000 Euro in consumables) for each condition tested. Direct sequencing can be done, but needs a different platform and analysis pipeline.

Regards

Gernot -- Gernot Gloeckner Genome Analysis Leibniz Institute for Age Research - Fritz-Lipmann-Institute e.V.

on 5/12/07 15:08, Jonathan Chubb at J.Chubb@dundee.ac.uk wrote: Dear All, I wondered if this might be the opportunity to ask to what extent anyone knows whether direct sequencing is feasible for AT rich genomes. Judging by the amount of whole yeast genomes that have been rapidly completed by this new method, and the attitudes of many chromatin biologists that this will fast supercede microarray technology for both expression and ChIP-CHIP studies- perhaps it would be worth jumping straight onto this? Regards, Jonathan

Dr. Jonathan Chubb Division of Cell and Developmental Biology School of Life Sciences University of Dundee Dow Street

on 4/12/07 12:14, Ludwig Eichinger at ludwig.eichinger@uni-koeln.de wrote:

Dear Thierry,

I agree with you that it would be of advantage to have a standardized genome-wide microarray chip for Dicty. However, I don't think Affymetrix is the best choice. I find it too expensive and I do not like being dependent on one company for the chip and downstream analyses. I also think that an array with 10100 features is not sufficient which would increase the costs.

An alternative would be an oligonukleotide array covering all genes and also including controls. The oligos (70-mers, amino-modified) for such an array with e.g. 13,000 features from Operon would cost around 150,000 Euro. The delivery can be in up to four normalised aliquots in 384 well plates without additional costs. This means that identical arrays could be spotted in e.g. USA, UK, Japan and Germany. The costs for such arrays would be low, presumably below 50,- Euro per array. In addition, one could make an agreement with Operon so that the oligoset can also be ordered from them (they sell oligosets for a number of organisms) if people wish to produce their own array. Currently I do not know the costs of follow-up orders for the oligo set.

If, let's say, four groups would be interested to share the costs for such an array, it might be more feasible to convince funding agencies to provide the money than if a single lab asks for the complete sum. Anyone interested in a joint effort?

Best wishes

Ludwig

on 4/12/07 20:53, Ted Cox at ecox@Princeton.EDU wrote:

Dear All,

I'm told by our local Genomics Brain Trust that Agilent arrays cost ~$450 for ~250,000 features. This seems about the cost/chip for Affy chips, minus the cost of a mask--whose design features will in any case change with time and experience. I'm also told by the heavy users that Agilent oligos (~70 mer's) give the cleanest results. The big advantage of Agilent is that one simply sends a file and they ship back the arrays. Perhaps, then, one simple goal might be for all of us to agree on 250,000 features for a common file? This could be modified with experience and no overhead costs.

But I certainly agree with Thierry that a resource with a common set of features would place all our data on the same landscape--so much now depends on individual skills using various cDNA collection whose inserts are of variable lengths, GC-contents, and so on.

Ted

Edward Cox Professor of Molecular Biology Edwin Grant Conklin Professor of Biology 333 Moffett Lab Phone: 609-258-3856 Lab Phone: 609-258-3571 Fax: 609 258 1343

on 3/12/07 19:13, wloomis@ucsd.edu at wloomis@ucsd.edu wrote:

Thierry,

As one of those who has printed and used a lot of microarrays, I was interested by your efforts to get Afffymetrix to build a whole genome array for community use. I also considered such an effort several years ago but then I ran out of money for such work. Moreover, Negin retired and I did not have the energy to start up again. So I closed up the microarray work in my lab and gave away the equipment. Actually, it was a relief to stop hassling over microarray data after 6 years of it. Basically, I feel that I got most of what I wanted from microarrays. A lot of genes and ideas came out of it. Now and then I think of an experiment where it would be nice to have microarray data. In fact, I am collaborating with Gadi on one. He still prints his own Dicty cDNA chips etc.

In many cases qRT-PCR can give the essential data and it is more quantitatively reliable. There is also the technique of pyro-sequencing that may replace microarrays. Directly sequencing >100,000 mRNAs in a single SOLiD or Selexa run and then counting up the number of times the same gene was encountered is a more direct way to see the transcriptional profile. I haven't looked into the cost of this approach but it holds a lot of promise now that the genome is completed. The bottom line to your e-mail is that I doubt I would order any arrays at this time. However, I wish you luck in rounding up a group that is interested.

What are your direct goals with this technology?

Bill