In late February, a month after Affymetrix drew away the curtain on its new U133 human genome arrays, which fit 45,000 unique probe sets representing the entire draft sequence of the human genome on two chips, BioArray News asked a selection of users how the chips stacked up compared to the previous generation U95v2 chips. While several freely delivered their opinions, most had barely begun to take their new toys out of the box.
Last week, BioArray News again contacted users, this time from a total of 30 different institutions, to inquire about their results with the chips as well as their thoughts on related products such as NetAffx, Affymetrix’s online microarray information database, and the Microarray Suite version 5.0.
U133 Users Say it Works
For the U133 chips, the general response was overwhelmingly positive. Users said that the chips offered little if any problems and offered highly reproducible results that either equaled or exceeded the results from the previous-generation U95 chips. Some thought this showed how Affymetrix has benefited from the acquisition of bioinformatics company Neomorphic, which has developed algorithms for selecting probes to be put on Affymetrix chips.
“The data [from U133 chips] makes more sense, it’s more reproducible,” said Gregory Khitrov, director of the gene array facility at Rockefeller University in New York. “It seems like the oligos they picked do a better job. The p-values are a lot tighter and the [coefficient of variation] is lower.”
While Mark Geraci of the University of Colorado Health Sciences Center recalled initially receiving a chip lot with a resolution problem, “Affymetrix replaced them all,” he said. Geraci has since been satisfied with these chips.
Others said they were happy with the U133, but did not find the quality much different than the U95. “Both the U95 and U133 have good, consistent quality controls,” said Laura Reid, director of genomic sciences at Expression Analysis, a Durham, NC-based company that runs Affymetrix array experiments for 39 different commercial, government, and academic customers. Reid did, however, note that “the total cost is less with U133.”
Holdouts for U95
Despite the obvious cost advantages of using U133 chips, a surprising number of users responded that they are still using U95v2 arrays. The reason: A significant number are performing long-term experiments with U95s and do not want to switch midstream.
“I encourage [researchers] to finish up a project with the U95v2 arrays because the data will not be comparable” to the U133, said Khitrov, echoing this view.
Reid said that about a third of human chip users at Expression Analysis still use the U95 arrays, and she has even seen a few who are still using the Hu6800 chip, the generation previous to the U95. “And with the mouse chips, we get people who are still using the 11K, who have not made the transition to the U74,” the current chip. “If they started the project with the 11K, they finish the project with the 11K.” However, she added that all new users of human arrays are encouraged to use U133 chips.
On the software side, Affymetrix earned rave reviews for NetAffx, but its Microarray Data Suite 5.0 garnered less praise. Some users even said that Microarray Data Suite 4.0 works better for certain applications (see Lab Report, p. 9). Others said that it was a good idea to re-run chip data with 5.0 that had been run with 4.0 within the same experiment. “In order to compare across the two analysis [methods], you have to reanalyze the stored arrays,” said Geraci.
Affy vs. Motorola, Agilent
Winston Kuo, a biomedical research fellow at Harvard Medical School, has begun a study comparing results of different human genome microarrays, including U133 and U95 chips, cDNA chips produced by Agilent, and Motorola’s CodeLink human chips. He used two breast cancer cell lines on all of the chips, and performed two replicates for each chip. In his preliminary results, he found that both U95 and U133 had comparably high rates of reproducibility and sensitivity. “Both are fine for any lab to use,” he said. However, Kuo has found that Motorola’s arrays have a higher specificity, in other words a lower number of false positives. Agilent’s cDNA arrays, on the other hand, had a lower specificity and sensitivity, but he said there were additional problems with these chips and they had to be re-run.
Kuo is also currently doing a larger scale cross-platform analysis of mouse gene expression tools, including Affymetrix, cDNA, 70-mer self-spotted oligo arrays, and serial analysis of gene expression, or SAGE technology, for a consortium of five research institutions. The consortium, which includes Harvard, Yale, Johns Hopkins, NIH, and Albert Einstein Medical School, wants to find out “whether errors occur” in different modes of gene expression analysis “and how to improve upon these methods.” In a separate study, Kuo said, Harvard is trying to establish a gold standard for gene expression technology and methods.
Additionally, Kuo is working on algorithms to translate gene expression data between Affymetrix and cDNA platforms. He is looking to convert cDNA Cy3/Cy5 ratios to absolute numbers, and convert Affymetrix absent/present calls to ratios, to figure out equivalency measures through which one form of data can be translated to another.
Meanwhile, many Affymetrix users are anxiously awaiting the U133-like updates for Affymetrix’s other products, which it has promised in the coming months. “We look forward to when they produce a similar product for the mouse and rat genome,” said Reid.