Entering edit mode
7.8 years ago
Learner
▴
280
I have done a SILAC experiment and I have identified many proteins now I would like to know the possibilities to identify complexes. One way is to check for known complexes which I don't want to do and I know how to use databases to identify known complexes. What i want is to find possible complexes that are not known. Do you think it is possible to do this based on H/L ratios or intensities ? Is there any way to do this ?
You don't give enough information about your data/experiment.
Did you do pull-down experiments ? In this case, SILAC data is normally used to derive confidence measures for the interactions. Once you have high confidence interactions, you can then proceed to build a graph of identified interactions with proteins/genes as nodes (assuming you have enough connected baits to build a graph) and apply a clustering algorithms to the nodes.
If you didn't do pull-downs but have differential expressions, you could assume that proteins in a complex vary in the same way and use one of the strategies used with microarray data.
@Jean-Karim Heriche no i did not do any pull down experiment , but a duplex SILAC, actually i can map the identified proteins to database and find complexes. what I want is to find co-migrating proteins not based on database but rather H/L ratio , do you know anyway to do this ?
What do you mean by co-migrating proteins ? Did you perform some purification steps ? If you want to infer protein complexes from your data, you have to have some sort of signal at least marginally correlated with complexes e.g. in differential expression experiments that could be proteins that vary together. If you only have relative abundance in only one sample, I don't think you have enough information in your data. However, you could use this to weight a protein interaction graph derived from other sources and then process with clustering this weighted graph.
@Jean-Karim Heriche not anything extra as purification. no I don't have relative abundance in only one sample, I have obtained H/L ratios for 100 samples . how can i use these H/L ratios as weighted graph? as you know we cannot obtained such ratios for all identified proteins so I do have a lot of NA as well but I can get raid of them anyway.
Your objective is still unclear to me. Are you then interested in finding groups of proteins that have similar relative abundance across several samples ? Are the values comparable between samples (e.g. did you use spike-in SILAC ?) ?
To use abundance as edge weight, combine the values for the two nodes being connected.
@Jean-Karim Heriche actually I want to find those proteins that possibly make a complex (I did not see any paper for duplex SILAC) Yes the values are comparable between samples
@Jean-Karim Heriche you mean I combine H/L ratios (abundance) as edge weight, combine the values for the two nodes being connected what will happen to those those that they have not any H/L ratio? is there any way to find like theoretical complexes ?
Unless there's something special you're not mentioning about your samples, you simply don't have information on complexes in your data (unless you redefine what a complex is). Shotgun proteomics is about getting a list of parts (with relative abundances in the SILAC case) but that doesn't contain information about what is interacting with what unless you design the experiment to capture that information.
There is maybe one other thing you could try and that is supervised learning. You could train a classifier to recognize known protein interactions with your data and use this model to assign an "interaction" score to all protein pairs. I found that this was actually tried for microarray data, e.g. this paper. The resulting scores can be seen as weighted edges of an interaction graph which you can then cluster to find complexes.
Concerning the missing ratios, you need to decide what they represent. It could be that the protein is absent or maybe just below threshold in one case. If you still want to keep them, you could use a missing data imputation strategy, e.g. failed detections could be assigned the value of the threshold used for detection.
@Jean-Karim Heriche
Sorry I can't access the full text of this paper. Our library doesn't seem to have a subscription.
@Jean-Karim Heriche OK, thanks
OK. Got it. I'll read it sometime this evening.
The method in the paper relies on HPLC fractionation of the samples so there is information about protein interaction. The idea is that proteins that interact have similar elution profiles. This is better explained in this paper. So the question remains, do you have any information on protein interaction in your data ? If you did sample fractionation like in these papers then why not use the same approach for the analysis ?
@Jean-Karim Heriche Thanks, do you know any other method like this when we do sample fractionation ?