Our readers may enjoy this story.
About 2.5 years back, a high-profile group gave us few RNAseq libraries to analyze. They were analyzing the same libraries by mapping on the genome (Bowtie/Tophat/Cufflinks), and asked us to check, whether de novo assembly gave any additional transcript.
You probably remember how transcriptome analysis programs were in late-2010. Trinity was not published, and Oases was the top dog. However, Velvet itself was a huge memory hog and Oases needed even more RAM to run. We had to spend quite a bit of social capital to find an appropriate (high-RAM) computer and run the assembly. Finally, after begging a friend with high-RAM computer, we managed to get several Oases runs completed.
We gave the requesting group a set of novel transcripts not identified through genome alignment. They received a full report, but here is most relevant paragraph.
Pairwise BLAST Comparison between Velvet and Tophat Sets
This analysis focuses on 48 hr set. We chose a BLAST cutoff 1e-20.
A. Comparison of 17790 high-FPKM Tophat genes with 38408 Velvet genes
Among 38408 Velvet genes, 28555 (74%) were not present in the Tophat set.
Among 17790 Tophat genes, 1148 (6%) were not present in the Velvet set.
B. Comparison of 78135 Tophat genes with 38408 Velvet genes
Among 38408 Velvet genes, 7269 (19%) were not present in the Tophat set.
Among 78135 Tophat genes, 20643 (26%) were not present in the Velvet set.
The above results are encouraging, because Tophat-Cufflink utilized the already assembled genome, whereas Velvet did not have such an advantage. Still Velvet assembled almost all (94%) highly expressed genes identified by Tophat.
Shock and horror - those biologists told us that they are already in the process of writing a paper with ‘conclusion’ that de novo assembly would give nothing new, but
asked us to do the analysis so that they could claim in the paper that they did run the analysis. We were not ready to go with that, and then came the next set of surprises. Those eminent biologists told us that we must have done something wrong in the analysis, because, well, they ‘concluded’ in their soon-to-be-submitted paper that assembly approach would not give anything new. To be honest, they did try to give few explanations on why de novo assembly would do things wrong.
(i) “Who knows what pieces those assembly programs are connecting” [We had to do more analysis of assembled fragments to show that they usually fell on the same scaffold, which is highly unlikely in case of random junctions.]
(ii) “Maybe those extra pieces were all UTRs” [We found that about 1/3rd to 1/2 of potentially novel transcripts were reasonably close to other genes to be classified as UTRs].
However, the general process of exchange was utterly frustrating, because those guys made up their mind and were trying to find some way to show that Velvet was wrong. Also, they kept making suggestions that they did not need to back their assertions with proof, because they were well-known biologists. So, we offered - “Why not you present your results in the upcoming conference, and we present ours? Then we will let the general scientific community decide who was correct.”
At that point, they pulled out their final trump card.
“We are the owners of primary data. You cannot present anything without our permission.”
What is the opinion of our readers? Who is allowed to present analysis of large-scale biological data in similar situations?