Notes from the biomass

Everyone is using tiling arrays these days but the quality of results is still hard to judge even for the most experienced people in the field. Designs differ in type of the oligo density, oligo length, selection algorithms and experimental procedures. On top of that, several analysis procedures claim to be superior over each other. Eight groups performed independent evaluations of tiling arrays for the human genome from the major vendors (Affymetrix, NimbleGen and Agilent) and report their findings in Genome Research, on February 7th, 2008. As the scientists had no knowledge about the PCR products that were spiked in the sample, this is an blind evaluation and thus much more powerful than the typical validations using real time PCR.

The overall results are a sobering: only 50% of the sequences selected were consistently detected at10% FDR. This is not to be taken literally as the majority of the missed samples were present in low numbers only (1.25 to 4 fold) and we do not have good data for the true fold changes in these experiments.

All three vendors supported this study and employ a good number of the authors. In that light, it is no surprise that the study does not report major quality differences between the chips. They do report that Affymetrix arrays require more repetitions to reach the same quality of the results but are of lower price (conveniently tabulated in the paper). No method performs well in the low concentration regime, although Agilent and Nimblegen arrays look a little better at it. I certainly don’t want to imply that the affiliations biased the results of this study (this is the internet after all). On the contrary, this is a very useful collaboration between chip vendors and technology leaders in academia. But read it before the next sales person in the chip business knocks on your door or you want to build trust in a particular data set.