Two papers recently published in Nature describe the human proteome (1, 2). But the protein identification procedures in these have recently been seriously criticized (3). I believe this is an excellent time/opportunity for us all to (still) learn valuable lessons for obtaining accurate proteomics data with today's instruments. Would anyone like to share their approaches to getting the good data in MS based proteomics?
1) http://www.nature.com/nature/journal/v509/n7502/abs/nature13319.html
2) http://www.nature.com/nature/journal/v509/n7502/abs/nature13302.html
3) http://pubs.acs.org/doi/abs/10.1021/pr500572z
That's really interesting, thanks for posting. As you probably know, there has been much debate about PTM identification and site localisation, but it seems the issue of peptide/protein identification was a "finished" issue. I guess that is not the case. I had expected journals would start to demand people use at least two search algorithms for identification and then combine that into one set. As far as I know, this has not occurred.
There are certainly enough tools around that let you use many search algorithms and then combine the result, so I think this is a very good way to go. Not only do you get higher quality data, you also identify more peptides/proteins with this approach. It obviously takes a bit longer, but it is well worth the effort. An issue here is if you are doing quantification, whether it be at the MS or MS/MS level. It starts to get very difficult, if not impossible unless you have a tame bioinformatician at hand.
Hi Peter, thanks for the input. Which search algorithms do you like to combine (e.g. open source, commercial)? Its a good point that journals could be more strict in their criteria of what is a quality search. I come from a method validation background, and would have liked to see a fairly complex QC sample/reference sample that could be used to demonstrate accuracy and even carry-over etc.
I'm a big fan of open source software, and I love SearchGUI (easy multiple search algorithm approach) and PeptideShaker (to combine the searches). I tend to use Mascot (in house server), then MS-GF+ and X!Tandem. OMSSA provides some good matches, but it is very slow and doesn't cope with large datasets. I find the same problem with MS-Amanda. Both are 32-bit, I believe the developers are working on a 64-bit MS-Amanda version. And OMSSA is essentially dead.
OpenMS is very flexible, but I think you need to be a coder to really work with this software, it's just not user friendly, even with the TOPPAS interface.
On the commercial side we use Scaffold. It's great for certain applications. This doesn't make searching any easier, but will combine most search algorithms.
The idea of some standard is a good one, that at least shows that you are doing the basics correctly. I guess this would have to come from the journals demanding such a standard is required for certain datasets. There would have to be a threshold where the standard is required though. Maybe we could work with several types of standards, depending on how complex the samples.
https://code.google.com/p/searchgui/
https://code.google.com/p/peptide-shaker/
http://open-ms.sourceforge.net/
- Badges
- Science topic