Personal tools
You are here: Home Chat Discussion with Roland Krause about protein-protein interaction networks
Document Actions

Discussion with Roland Krause about protein-protein interaction networks

Roland Krause answered questions about his protein-protein interaction networks talk. This IRC discussion took place at the third day of the symposium (Wednesday, December 6, 2006, 16:00 CET).

The questions were answered by Roland Krause (rolandkrause), participants where

  • Malgosia Duszczyk (mGosia)
  • Grege Tyrelle (greg_tyrelle)
  • Egon Willighagen (egonw)
  • chaired by Konrad Förstner (konrad_foerstner)

[We started 12 minutes later.]

16:12 <@konrad_foerstner> I would like to welcome all logged in participants. Today Roland Krause who is leading the group of "Bioinformatics in Microbial Virulence"
16:12 <@konrad_foerstner>  which is part of the Max-Planck-Institute for Molecular Genetics and the Max-Planck-Institute for Infection Biology in Berlin will be online with us.
16:12 <@konrad_foerstner> I assume everybody in the chat room has watched his presentation about protein-protein interaction networks.
16:12 <@konrad_foerstner> He will now answer your questions regarding this topic.
16:13 <@konrad_foerstner> Thank you Roland.
16:13 <@konrad_foerstner> Are there any questions?
16:13 < rolandkrause> thanks Konrad 
16:13 < rolandkrause> and thank you for the invitation
16:13 < mGosia> I have to say I only very quickly looked through the presentation
16:14 <@konrad_foerstner> okay.
16:14 <@konrad_foerstner> Any questions?
16:14 < rolandkrause> don't worry, just ask
16:14 < egonw> mGosia: same here, and I am not even into protein interaction much (more into small molecules)
16:14 < greg_tyrelle> Roland,
16:14 < mGosia> What I understood is that you want to integrate bioinformatics methods to make TAP purification for complexes more reliable
16:14 < greg_tyrelle> how extensive are transient interactions ?
16:15 < greg_tyrelle> i.e. how many are you missing ?
16:15 < rolandkrause> oops, two questions - I'll take mGosia first if you allow
16:15 <@konrad_foerstner> yes, choose yourself, please
16:15 < rolandkrause> Yes, well, actually, several people have done that
16:15 < rolandkrause> (integrating PPI data)
16:16 < rolandkrause> and I think that the overlap on one end is limited
16:16 < rolandkrause> and on the other functionally not relevant
16:16 < rolandkrause> a metabolic pathway does not need to be reflected in physical interactions of the proteins
16:17 < rolandkrause> questions answered?
16:17 < mGosia> Sort of...
16:17 < rolandkrause> if you give more precise questions, I am happy to elaborate
16:18 <@konrad_foerstner> otherwise we come back to it later and go on with the next question
16:18 < mGosia> OK, have to think about it :-)
16:19 < rolandkrause> OK
16:19 < rolandkrause> so, greg_tyrelle mentioned transient interactions
16:19 < rolandkrause> did I get you right in that you were interested in the number of interactions of that type?
16:20 < greg_tyrelle> a feel for the extent, are there numbers ?
16:20 < greg_tyrelle> I understand TAP will only pull down stable interactions ?
16:20 < rolandkrause> depends on your definition and that is the point
16:21 < rolandkrause> if you define transient in biochemically weak, yes
16:21 < rolandkrause> if you apply other measures such as stable through cell cycle - no
16:21 < rolandkrause> my point in the presentation was on that distinction
16:22 < greg_tyrelle> I was thinking more in terms of the second definition
16:23 < rolandkrause> well, there is an interesting work fro Lars jensen in my references
which goes into depth on that measure
16:23 < rolandkrause> the complexes that they build on can be purified by TAP, at least in principle
16:24 < rolandkrause> however, quantification might be difficult
16:24 < greg_tyrelle> how far along is the biacore work you mentioned in the presentation?
16:25 < rolandkrause> what  do you mean - it is a pretty established method used in many labs but you can't use it on complexes
16:26 < egonw> rolandkrause: (when you're done with the current question...) I am a chemist by background, so quite a novice in this area... but how do you actually get from MS data to the identity of the proteins in one complex? each MS spectrum is for one complex? so gives one peak per complex component in the spectrum?
16:26 < rolandkrause> we looked into getting KDs (interaction binding constants) but there's hardly good data
16:26 < mGosia> Have you tried NMR ;-)
16:27 < rolandkrause> OK, I take egonw's basic questions and we can come back to transient interactions later
16:27 -!- laketa [***] has joined #embl_online_phd_symposium
16:27 < greg_tyrelle> sure
16:27 < rolandkrause> No, you separate the complex on a gel or LC, trypsinize the proteins
 and use a MALDI to identify the fragments
16:28 < rolandkrause>
16:28 < egonw> ok, so per complex you get a mixture of fragments, from which you get one or more protein hits... and if more, it was a complex?
16:28 < rolandkrause> is a good review to get started with pictures
16:29 < egonw> ah, thanx for the link
16:29 < rolandkrause> no, the peptide fragments have a unique hit in the protein database of the organism
16:30 < greg_tyrelle> thank you roland for taking the time to answer my lest than specific questions :) I must leave, it's late here
16:30 <@konrad_foerstner> thanks greg for participating
16:30 < greg_tyrelle> it was worth it, see you all
16:30 -!- greg_tyrelle [***] has quit ["leaving"]
16:31 < egonw> ok, I misread the 'separate the complex'... I took that as 'separate the complex mixture'... got that now
16:32 < rolandkrause> in principle you could try the complete proteins from the complex mixture but that method has its limits
16:32 <@konrad_foerstner> I have another basic technical question: Are there known cases in which the fusion of the TAP-tag influences the structure of the bait, which leads to an different interaction behaviour (so produces non-natural results)?
16:32 < egonw> how many complexes are commonly found in a cell, urine sample, what ever the source is?
16:33 < egonw> oh, that crossed...
16:33 <@konrad_foerstner> no problem
16:33 < rolandkrause> Konrad first
16:33 < rolandkrause> basic questions
16:33 < rolandkrause> yes, there are known examples where the tag interferes with the protein
16:34 <@konrad_foerstner> how big is that fraction?
16:34 < rolandkrause> Elements of the proteasome, CCT chaperones
16:34 < rolandkrause> < 10 %
16:34 <@konrad_foerstner> okay. thank you.
16:34 < mGosia> And how do you know this is the case?
16:34 < rolandkrause> problem is that there it is difficult to show on a large scale
16:34 < rolandkrause> but you can modify histones with much larger tags and have living c. elegans
16:35 < rolandkrause> mGosia: these proteins have a known functional site at the C-terminus
16:35 <@konrad_foerstner> okay, the living phenotype tells roughly the interaction takes place as it should do
16:36 < mGosia> You cannot attach the TAP at the N-term?
16:36 < rolandkrause> you can
16:36 < mGosia> OK
16:36 < rolandkrause> it is a little more tricky (don't ask about the details) - for another large screen that would be nice to have
16:36 < mGosia> I won't ask :)
16:37 <@konrad_foerstner> Any other questions?
16:37 < rolandkrause> Good, we go back to egonw regarding the number of complexes
16:37 <@konrad_foerstner> good
16:37  * egonw is proud to have asked a non-basic question :)
16:38 < rolandkrause> so, the number of complexes in the yeast cell is 500 - 800
16:39 < rolandkrause> In principle, you need to define how you count variant complexes and
 megacomplexes - is the ribosome one, two or three (subunits + holocomplex) complexes?
16:39 < mGosia> And the estimation in human?
16:39 < mGosia> Compared to that?
16:40  * egonw is trying to look up the number of genes of yeast
16:40 < rolandkrause> OK, here you need to define your level of complexes
16:40 < rolandkrause> 5500 genes in yeast
16:41 < egonw> does that mean that a lot of proteins (assuming an general one-gene-one-protein scheme) do not form complexes?
16:41 < rolandkrause> problem is that you have much more complexity due to protein family expansion, so you cannot multiply the number easily
16:41 < rolandkrause> yes
16:41 < egonw> mostly enzymes?
16:41 < rolandkrause> about 40 - 60% are not in complexes
16:41 < rolandkrause> many enzymes
16:42 -!- lysin [***] has joined #embl_online_phd_symposium
16:42 < rolandkrause> but there are a lot of proteins that we have no readings for - no functional data, we don't know whether are expressed in the lab ever
16:42 < egonw> ok, good point
16:43 < mGosia> What about the transient, as in biochemically weak, complexes?
16:43 < rolandkrause> I still try to avoid to answer the questions about complexes in human
16:44 < mGosia> That's OK :) So the transient, as in biochemically weak, complexes in yeast
16:44 < rolandkrause> mGosia: well, they exist, they are difficult to find
16:44 < rolandkrause> with most methods at hand
16:45 < rolandkrause> some people argue that synthetic lethals or yeast-two hybrid can identify them
16:45 < rolandkrause> which I doubt - not because the methods are bad in any sense but because they measure different aspects of interactions
16:46 < mGosia> Or e.g. with Biacore, or NMR, but those usually can only handle 2 proteins at a time, so no complexes :(
16:46 < rolandkrause> yep
16:46 < mGosia> Bummer :)
16:47 < rolandkrause> yes, but you also have to kiss the interactome good bye
16:47 < rolandkrause> it is a term for all interactions but every new method will find dif
ferent ones
16:47 < rolandkrause> and fail to reproduce many of the established ones
16:48 <@konrad_foerstner> I have a question regarding reproducibility. Do the results of the mentioned Saccharomyces cerevisiae analyses (Gavin et al. and Krogan et. at.) confirmed each other?
16:48 < egonw> rolandkrause: a bioinfo question: how do you find interesting regions in the interaction graph? Do you use clique detection, or something else... ?
16:49 < egonw> (arg... another cross... sorry :)
16:49 < rolandkrause> konrad first :-)
16:49 < egonw> sure
16:49 <@konrad_foerstner> no problem :)
16:49 < rolandkrause> no, they do not reproduce each other too well on a superficial level
16:49 <@konrad_foerstner> but they are comparable in general?
16:49 < rolandkrause> which is surprising given that both use the same method
16:50 <@konrad_foerstner> this sounds fishy
16:50 <@konrad_foerstner> how does the community handle that?
16:50 < rolandkrause> what I have observed is that both methods reproduce the known complexes fairly and agree in that areas
16:50 < rolandkrause> but for the potentially novel data, the disagreement is much larger
16:51 <@konrad_foerstner> so there is no real gain of knowledge
16:51 <@konrad_foerstner> ?
16:51 < rolandkrause> the problem of integrating the data is that not the same basic experimental information
16:51 < rolandkrause> on the contrary
16:52 < rolandkrause> there are plenty of novel complexes, mostly small and a substantial number of expansions
16:52 < rolandkrause> of known ones
16:53 < rolandkrause> but this is the first time we get a complete view and while several components and complexes were identified previously, we have no birds eye on the data
16:53 < rolandkrause> the small scale experimental data has the same problem with contaminants
16:54 <@konrad_foerstner> okay.
16:54 < egonw> is that similar to the microarray problems?
16:54 < egonw> where obvious things can be found in the data
16:55 < egonw> but the interesting things are hidden in the experimental error?
16:55 < egonw> (roughly speaking, and generalizing a bit here...)
16:55 < rolandkrause> that trend you have with all high-throughput data
16:55 < rolandkrause> so, yes but not as bad, I would say
16:55 < rolandkrause> with microarray data, you don't have a gold standard
16:55 < egonw> good... follow up question...
16:56 < rolandkrause> here we have literature curated data
16:56 < egonw> if experimental error is a problem, are you using duplicates/triplos to cancel out errors?
16:56 < egonw> of is that, like with microarray data, to expensive?
16:57 < egonw> to -> often too
16:57 < rolandkrause> you have to balance costs
16:57 < egonw> yes, I understand...
16:57 < rolandkrause> We have a first pass complex interaction network
16:58 < rolandkrause> and it would be nice to bring to a better level by repeating the experiments, actually one of my main points
16:59 < rolandkrause> generally, experiments by other methods are preferred to repetitions in biology
16:59 <@konrad_foerstner> In your talk you proposed a community effort scheme for this. I guess this is not yet going on. How many lab are working in that field and need to be brought together for this?
17:00 < rolandkrause> dunno
17:01 < rolandkrause> I am in touch with some labs but I guess it would require a major player to bring this forward
17:01 <@konrad_foerstner> which role play companies in the community?
17:01 < rolandkrause> Also, I have to admit that I fear the amount of work that comes with starting something like this
17:02 < rolandkrause> the role of companies is limited - and must be because there is very little money in this kind of basic science
17:02 <@konrad_foerstner> okay, too far away from any application at the moment.
17:03 < mGosia> Are there any other methods available to question the 'novel' interaction data you get out
17:03 < rolandkrause> yes and no - cellzome is pursuing protein-protein interaction in human
17:03 < rolandkrause> but they don't need to define things for the community but with their clients
17:04 < rolandkrause> mGosia: Yes
17:04 < mGosia> I guess at smaller scale you can look at colocolization etc but at a larger scale?
17:04 < rolandkrause> All of them
17:04 < rolandkrause> ;-)
17:04 < rolandkrause> you are right - we don't have much of an idea for several completely
 novel proteins
17:05 < rolandkrause> but I know several complexes that were identified via interactions in the Gavin screen from 2002
17:05 < rolandkrause> I don't believe it is a smart approach to go through the complexes one by one
17:06 < rolandkrause> Instead, the data is taken from SGD or other sites and specialists in one field or another pick it up
17:07 <@konrad_foerstner> Further questions?
17:08 < mGosia> nein :)
17:08 <@konrad_foerstner> I would like to ask you about the problems analysing human complexes
17:09 < mGosia> Thanks a lot roland, gotta get back to the lab!
17:09 <@konrad_foerstner> bye mGosia
17:09 < rolandkrause> mGosia: thanks for tuning in. konrad_foerstner; shoot
17:09 -!- mGosia [***] has quit []
17:10 <@konrad_foerstner> What are the technical hurdles for analysing the (more interesting) human interactions
17:10 <@konrad_foerstner> ?
17:11 < rolandkrause> the biggest problem from my point of view is that there is no homologous recombination as in yeast
17:11 < rolandkrause> so, getting a fusion protein is more difficult
17:11 < rolandkrause> and you can't use the natural promotor
17:12 < rolandkrause> so, typically you have to overexpress the protein and get countless artifacts
17:12 <@konrad_foerstner> okay, and doing this large scale will cost a fortune.
17:13 <@konrad_foerstner> Are there further questions?
17:13 <@konrad_foerstner> This seems to be not the case.
17:13 <@konrad_foerstner> I want to thank Roland and all of you.
17:14 <@konrad_foerstner> +++ THE END +++

Powered by Plone CMS, the Open Source Content Management System