[super0925]

"2'-5'-oligoadenylate synthetase" is an example of a gene name, yes. I also can't get bioDBnet to work with that example, but I've never used the site before. As I suggested earlier, just parse the annotation file you're using, since it apparently have both.

Sorry I haven't got your idea. Do you mean once I get the gene name list , like OAS1, PTGDR, etc. If I want to transfer them to their full name (or as you said the example of gene name) to "2'-5'-oligoadenylate synthetase","prostaglandin D2 receptor", I need to Google it?
Could　ｔｈｅ　annotation file e.g. GTF file have these full name? I only remember it has the Ensembl ID and hence I need to transfer the ID to gene names (e.g. OAS1) by online converter tools. But for how to transfer them to the full name, I am still confused.
Thank you!

[dpryan]

"2'-5'-oligoadenylate synthetase" is an example of a gene name, yes. I also can't get bioDBnet to work with that example, but I've never used the site before. As I suggested earlier, just parse the annotation file you're using, since it apparently has both.

[super0925]

Perhaps the uniprot IDs haven't been merged into biomart yet. Since you have an annotation file, the conversions are contained in it. So just load it into R with GenomicRanges and then get the conversions from the mcols() of the resulting GRanges object.

A "2'-5'-oligoadenylate synthetase" would add ATP to oligos (if you google this, you'll find that apparently this particular method ends up activating ribonucleases and leading to RNA degradation).

Thank you! But bioDBnet (http://biodbnet.abcc.ncifcrf.gov/db/db2db.php) has merged the uniprot, however it seems not working. Could you please try it? I am wondering if I selected the wrong catalog (e.g. gene symbol, gene ID...). Because if I input the gene symbol 'OAS1 ', and the Outputs select the 'UniProt Accession', I don't find the 'Q865A2' in the result list, and hence, if I input 'UniProt Accession' is 'Q865A2', outputs is null. It is strange!

BTW， how to define the "2'-5'-oligoadenylate synthetase" for 'OAS1' , or "prostaglandin D2 receptor" for 'PTGDR'? Are these long name called "Full gene name"? Could I use Biomart or bioDBnet or R packages you mentioned to transfer the Ensembl ID to this so called 'Full gene name'? Or the only way is to Google them one by one after I got the list?
Thank you!

[dpryan]

Perhaps the uniprot IDs haven't been merged into biomart yet. Since you have an annotation file, the conversions are contained in it. So just load it into R with GenomicRanges and then get the conversions from the mcols() of the resulting GRanges object.

A "2'-5'-oligoadenylate synthetase" would add ATP to oligos (if you google this, you'll find that apparently this particular method ends up activating ribonucleases and leading to RNA degradation).

[super0925]

org.Bt.eg.db from Bioconductor. It has a uniprot to ensembl conversion map. BTW, you can also script biomart with the biomaRt package in R (this is very useful for things like ID conversions and fetching sequence information).

I have two questions now:
1. I have tried the Biomart(http://www.ensembl.org/biomart/martview/) on Q865A2_BOVIN but I didn't find the result...
2. I also try bioDBnet (http://biodbnet.abcc.ncifcrf.gov/db/db2dbRes.php) and it is not work as well.
Could you please try it? (may take you 1 minute) I think may be I wrongly choose the catalog (e.g. UnigeneID, Gene ID) of the input or output.

The result of Q865A2_BOVIN is OAS1 （please see (http://www.uniprot.org/uniprot/Q865A2), btw what is '2'-5'-oligoadenylate synthetase', could it be transfer to from Biomart or bioDBnet?

Thank you!

[dpryan]

org.Bt.eg.db from Bioconductor. It has a uniprot to ensembl conversion map. BTW, you can also script biomart with the biomaRt package in R (this is very useful for things like ID conversions and fetching sequence information).

[super0925]

Q865A2_BOVIN looks like a uniprot ID (with a "_BOVIN" suffix, perhaps), so just convert that with Biomart or R (there's probably a bovine annotation package). BTW, this is usually not an issue since the IDs output by cufflinks and those used by htseq-count can be the same (they're both sourced from the annotation file you provide, so you can just tell htseq-count to count according to the same ID).

Which R package could do this? I know how to convert the Ensembl gene ID list to gene names list on Biomart website http://www.ensembl.org/biomart/martview/ ,but it is manually (I mean I need to copy and pasted one list by one list), not automatically.

[dpryan]

Hi how do you uniform the gene name between different methods (Texudo vs Htseq+edgeR)?
Now I am doing RNA-Seq on the bovine, my reference genome use Ensembl database.
As you know, the Texudo pipeline: Tophat2-Cufflinks-Cuffmerge-Cuffdiff.
After the Cuffmerge and get the merged.gtf which could give you gene name output only but not the Ensembl ID. However, the gene name of Cuffdiff, some are like "Q865A2_BOVIN", which made me confused, cause it is not official gene name.
Is there any methods to uniform gene names (whatsoever Ensembl ID, gene name or UCSC ID ,.etc.) across different methods?
I think I could do Tophat2-Cuffdiff without Cufflinks+Cuffmerge to do this pipeline. However, I am not sure I am right and is there any other way to uniform them?
Thank you!

Q865A2_BOVIN looks like a uniprot ID (with a "_BOVIN" suffix, perhaps), so just convert that with Biomart or R (there's probably a bovine annotation package). BTW, this is usually not an issue since the IDs output by cufflinks and those used by htseq-count can be the same (they're both sourced from the annotation file you provide, so you can just tell htseq-count to count according to the same ID).

[super0925]

Hi Devon and Zapages,
Have you seen my questions on post #92 ? Thank you!
I am confused on how to uniform the gene name or ID between Tuxedo and count based pipelines. The different gene annotation file is not convenient especially I don't know how to deal with sth like "Q865A2_BOVIN"

[dpryan]

CPM is basically a non-length-normalized variant of RPKM, so it has many of the same pros and cons. In an ideal world, RPKM should be the most accurate of the three for many things, since if you have differences in major isoform length between groups then it can better represent that. Having said that, it's a bit more difficult to calculate accurately and then use well statistically (look at the evolution of cufflinks for an example). The normalized counts from DESeq2 (or edgeR) are mostly used for visualization and not much more. In most cases, there's not one "best" thing. These are all metrics that are more/less useful for a variety of things. If CPM has better predictive power in your samples then I wouldn't argue much with that. All the theoretical discussions in the world have to take a back seat to what validates experimentally.

[sindrle]

Originally Posted by dpryan
....FPKM/RPKM values have some downsides in that they lead you to believe that you've properly corrected for library size differences, when you actually haven't robustly.
...
Some people mistakenly believe that the count-based method is "biased" toward longer genes. This is incorrect since that's not a real bias (in the common rather than technical meaning), that's a benefit. Etc.
...
.
Could you please elaborte on these two points please?

When Im analysing expression profiles I export one table with RPKM, one with CPM and also one with normalised counts from DEseq2.

However I don't know which one is the best? So far I have used CPM values, because the have correlated best with RT-PCR delta CT values. Also with corresponding protein values.

How do these three methods relate? What are pros and cons?

[super0925]

Biological replicates are the most important. In an ideal world, you'd do a pilot experiment and then use a tool like scotty to determine how many replicates you would need for the power you want.

Hi how do you uniform the gene name between different methods (Texudo vs Htseq+edgeR)?
Now I am doing RNA-Seq on the bovine, my reference genome use Ensembl database.
As you know, the Texudo pipeline: Tophat2-Cufflinks-Cuffmerge-Cuffdiff.
After the Cuffmerge and get the merged.gtf which could give you gene name output only but not the Ensembl ID. However, the gene name of Cuffdiff, some are like "Q865A2_BOVIN", which made me confused, cause it is not official gene name.
Is there any methods to uniform gene names (whatsoever Ensembl ID, gene name or UCSC ID ,.etc.) across different methods?
I think I could do Tophat2-Cuffdiff without Cufflinks+Cuffmerge to do this pipeline. However, I am not sure I am right and is there any other way to uniform them?
Thank you!

[dpryan]

Biological replicates are the most important. In an ideal world, you'd do a pilot experiment and then use a tool like scotty to determine how many replicates you would need for the power you want.

[super0925]

You don't do a DE analysis without replicates. The best you can do is look at the ranked fold-changes (have a look at the GFold package, which tries to do this in a somewhat more useful way). Personally, I wouldn't waste my time without good reason.

Hi Devon I remember that last time you are talking biological or technological replicates.
I think that how many minimum or maximum replicates in library preparation is based on the experiment design, but could we decide it from Deseq/edgeR results (e.g. outliers or expression levels) or we only decide it just based on the experience ? because as you know, the more replicates requires the more money for sequencing.
My understanding, the more replicates is more useful theoretically, am I right?
In library preparation, what is the impact of reads length and depth? are they the more the better ? and if the depth is more important than length?
Thank you!

[pecanton]

It's generally a better idea to use the variance stabilized data when making heatmaps. How you want to filter things is up to you. Randomly subsampling things, using only the DE genes, or filtering based on variance are probably the most common choices. The method to choose depends on what you hope to gain. I personally do clustering mostly for quality control, since it's largely not that informative otherwise.

BTW, you might be interested in using WGCNA, which can be thought of as using clustering to do more rigourous groupings of genes/probes. It's an R package available on CRAN and is probably not the most user-friendly package ever, but the methods might be really appealing for the questions you have in mind.

Regarding centering, usually you'll want to center the values. After that there's no perfect answer to things. Lior Pachter wrote a blog post a while back regarding making heatmaps that you should read. I also wrote a bit about this in an answer a while back over on biostars that you might find helpful. The short answer is that there is usually no single best answer.

That post by lior Pachter was really informative. Thank you for that. I also read your post at Biostars. I think I may settle on filtering the variance stabilized matrix by the DE genes and NOT using the euclidian distance between expression data. However, I'm wondering how it will look. In the matrix I have 15 columns, 3 replicates for each of the 4 time points and the control condition. The table after the DE analysis and the log 2 changes only has 4, since DEseq2 has already evaluated the replicates into one result, and it implicitly has the comparison to the control condition. I'd think longer vectors for each gene would allow better clustering through any algorithm, though I still get the impression that it's for debate which one is more accurate, just that one should shy away from those that cluster mainly on min or max values.