Don't know if this is helpful, but the Expression Analysis company
Company Website:
http://www.expressionanalysis.com/RNASEQ
offers, as part of its Bioinformatic support:
"c. .CEL file generation based on transcriptome quantitation compatible with the Human HG-U133_Plus_2 array, the Rat230_2 array, or the Mouse430_2 array"
I'm sorry, but what is the purpose of doing so? While both are measures of RNA abundance, I'm sure they are quite different data from a statistical standpoint, and the methods used to analyze them (separately or jointly) should account for those differences.
Max, clustering methods for gene expression patterns are highly developed from chip data. These have resulted in clinical level descriptions of subsets in breast cancer and many lymphomas. You are truly disregarding a great deal of formidable work that has already been done if not considering some kind of transformation between the two data types. My 30+ years of experience tells me this is true.. Joy, although it's obvious that two different platforms aren't directly comparable, the point is that it should be possible to formulate a correlation that will work for defining, for example, triple negative Breast cancer from Luminal B (for example).
Send me your RNA sequences: I will make W-curves out of them and look for their 3-D Informational patterns
Bert, converting the RNASeq data to the same format as chip data is not a statistically valid transformation of the data, and the more highly developed the clustering methods for chip data are, the more they will depend on the fact that they are in fact clustering chip data, not some other kind of data with far different statistical properties. Rather than ignoring the chip data approaches, I'm suggesting you should respect them and only use them on the data for which they were engineered. The right way to leverage the work that has been done on clustering chip data is to transfer the approaches to RNASeq-specific approaches that are as specialized to RNASeq data as the mature chip data approaches are specialized to chip data.
A car designed to run on unleaded gasoline does not run efficiently on diesel fuel.
Joy, this is a smart answer but needs principal components analysis and coupling with somatic and germ line genetics to be truly useful. I think there is a real wrongheaded belief in the community that RNAseq will supplant expression chip data. Not this year; and very possibly not for the next five..., no matter how enthusiastically the proponents of RNAseq like it.
The faddish history of technology development in the biosciences supports the view that, just as we perfect one technology (microarrays have come a long way in the past few years), we abandon it. Just plain silly and more related to the trendiness of hula-hoop than most scientists and funders are willing to admit.
Bert, I couldn't agree more. I simply see RNASeq and microarrays as different kinds of data, for which normalization and other techniques for wringing as much information out of each kind of data generally hurt, rather than help, when applied to data generated in a different type of experiment. MeV sounds like it takes the right approach: treat each kind of data as different even when analyzing them together.
I think there are some very useful answers in here - technology fads, yes, but we have to live with them if that's where the money goes - this will only get worse I expect, and yes we have to use the right tool for the job.
The dynamic range has been mentioned. Also, it's likely that a lot of genes will have the value 0 for their expression in RNA-Seq data, and it is not clear whether 0 means no expression or not. Also, array data seem to be generally expected to be normally distributed, whereas RNA-Seq data are not - there is roughly a consensus that read-count data are Poisson-distributed, but after normalisation, what then? I have seen bimodal RNA-Seq normalised data. This difference in distribution is a reason why we must be very careful before we use any array analysis tool for RNA-Seq analysis.
I will be having a look at the new MeV to see how they handle some of these issues. Meanwhile, is there a statistician in the room who can recommend an appropriate clustering method for data which was once roughly Poisson-distributed before it was normalised?
I have seen the work by J. C. Marioni, C. E. Mason, S. M. Mane, M. Stephens, and Y. Gilad. and I will say its a good piece in respect to genome study
By now I have looked at Marioni et al., and realize that they didn't do a complete job. There are many diseases with characteristic expression patterns now. One needs to use those for factor analysis in RNAseq and in chip data independently and then compare the differences. Liver v. Kidney appears a poor choice for the expenditure of big bucks in these austere times.
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology