Re: [Psidev-pi-dev] Cross-linking discussions at PSI meeting

[Robert C. <cha...@cg...>]

All,

Thank you Eilen for the notes on the session: it looks like you had a 
fairly productive discussion.  Sorry I could not be there: I had to 
teach yesterday.

Some comments:

ms-3 data: each spectrum is going to be identified as a linear peptide 
with a modification.  I believe Thermo are going to market the DSSO 
cross-linker, which produces two different potential modifications (the 
cross-linker can cleave in two different locations).  Hence, you would 
need to store which modification was found on which peptide, and then 
maintain the association between two MS3 spectra to infer the original 
cross-linked product.

Scores:  I think it is very important to allow storing of scores for 
each peptide separately, because as I have published, when you work with 
complex mixtures the reliability of your overall result is basically 
determined by how reliable the less confident of the two peptide 
identifications is (one of the two peptides is normally identified with 
at least an order of magnitude higher confidence than the other).

Post-processing:  I would assume this would just consist of another 
score: analogous to how people may use Peptide Prophet or Percolator to 
add another score on top of the original search engine result.

Dead-end/monolinks and Unimod: Several of these are already in there 
with the prefix ‘Xlink:…’.  I previously tried to have a discussion 
about whether dead-end modifications or true cross-links should use this 
prefix, but there was no agreement.  To some extent I think it boils 
down to whether Unimod should even try to capture true cross-link 
modifications.  If the answer is ‘no’, then we can use the Xlink: for 
the dead-end, and we just need to add a few more terms to Unimod.

Fragment ions:  There are plenty of regular b and y ions. Theoretically, 
half the fragments don’t contain the second peptide; in practice it is 
more than that because you mostly see the smaller fragments.  Also, 
software that identifies cross-links as linear peptides with mass 
modifications will report b and y ions containing the second peptide.

Cross-linking to DNA/RNA: I agree that this should be treated as regular 
peptide modifications analogous to glycosylation: no attempt is made to 
determine the sequence of the RNA/DNA, and in all these studies the 
RNA/DNA is digested down to only two or three residues, so there will be 
a manageable number of modification compositions that need to be listed.

DSS/BS3: PSI-MOD made the decision to make all modifications 
product-centered; e.g. oxidation (M) and oxidation (P); one is an 
oxidation reaction; the other is a differently synthesized amino acid 
(hydroxyproline).  The search engine software can mistake these two, so 
the ambiguity should be captured in the results.  I see no reason why 
someone couldn’t do a crosslinking experiment where they added both DSS 
and BS3 to try to get cross-linking of more types of proteins.  The 
results would not be able to distinguish which reagent formed each 
crosslink.  If one used reagent-centered names how would the results be 
reported?

Visualization of data: MS-Viewer could be used to display identification 
results, but we have no immediate plans to support mzIdentML 1.2 for 
cross-linking.  However, if a mzTab equivalent was produced we would 
work to quickly support it.

Lutz’s cross-link examples:  We have to remember mzIdentML is storing 
the results from a database search engine.  There is no software out 
there that tries to identify 4 and 5 (we have a hard enough time 
identifying 1!).  In the antibody example this is not going to be 
identified by MSMS analysis; this would be inferred purely by mass and 
knowing the protein sequence of the antibody you are studying.  It is 
unlikely that software would be able to identify 3 either, because if 
there are two cross-links then most peptide backbone cleavages will not 
produce a fragment.  We only should worry about 1 and 2.

Robert

On 4/20/2016 12:09 AM, Elien Vandermarliere wrote:
>
> Dear all
>
> In attachment is the summary of the session on crosslinking.
>
> Elien
>
> Elien Vandermarliere, Dr
>
> /VIB Medical Biotechnology Center, Ugent/
>
> /UGent faculteit Geneeskunde en Gezondheidswetenschappen/
>
> /Albert Baertsoenkaai 3/
>
> /B-9000 Gent/
>
> /BELGIUM/
>
> //
>
> /Tel. +32 9 264 9359/
>
> /Fax +32 9 264 9490/
>
> /www.ugent.be/
>
> /www.vib.be/
>
> *From:*Jones, Andy [mailto:And...@li...]
> *Sent:* dinsdag 19 april 2016 16:56
> *To:* tobias
> *Cc:* eug...@tu...; le...@im...; Sule 
> Yilmaz; ju...@ed...; Elien Vandermarliere; Lutz Fischer; Robert 
> Chalkley; Juan Antonio Vizcaino; Yasset Perez-Riverol; 
> psi...@li...
> *Subject:* Cross-linking discussions at PSI meeting
>
> Hi all,
>
> This email is addressed to those who participated in the XL session at 
> PSI today, plus Lutz and Robert (and plus Yasset for GitHub issues). 
> Please forward this on to others who should be consulted. I have also 
> copied to the PSI-PI list.
>
> The major outcomes from today are as follows:
>
> 1. General agreement that we are nearly there with the mzid 1.2 
> encoding of XL info.
> 2. Plan for Lutz to maintain (via mzIdentML GitHub for example), a 
> separate CV of crosslinker reagents, based off the attached sheet 
> (converted to CSV format, with unique IDs per row e.g. XL:0001, 
> XL:0002, more discussion needed to finalise format)
> 3. Some additions to mzid 1.2 to cover:
> - cases of combined evidence from multiple input spectra (L v H 
> isotopes; ETD+HCD; MS3 etc) to make an identification (not 
> post-processing but intrinsic to the ID mechanism)
> - Adding protein-level interaction evidence
> - Re-using additions already proposed in mzid 1.2 for mod or XL 
> localization ambiguity
> 4. No major interest at this stage to encode XL info in mztab, we may 
> do this at our side following same model as mzid 1.2
> 5. Plan for wider project over medium to long term to write a 
> cross-linking standards paper, including mzid 1.2 plus a minimum 
> reporting guidelines doc (Juri / Alexander to progress this), with 
> wider authorship
> 6. We will role these changes into mzid 1.2 specifications, and 
> publish that paper separately (submitted in the short term). For those 
> that contribute example files or major input of the specs, I will 
> invite to join the author list of that paper.
>
> I have started to write this up in the mzid 1.2 specification document 
> (attached here and committed to our GitHub) - see pages 20-23), and an 
> XML snippet for how the protein interaction part will look.
>
> ACTION items:
>
> - We discussed Lutz updating the xi export examples to include protein 
> interaction scores, following our proposed spec
> - Alexander and Eugen to update the examples exported from OpenMS and 
> add to the mzid GitHub (please email Yasset - cc'd to get write access 
> to the GitHub)
> - Ideally, would be great if the Edinburgh visualisation software 
> could read the mzid 1.2 files to see if all info is really covered
> - Andy to continue cleaning and updating specification document
>
> I would like to propose a conference call to progress this for *4pm 
> UK/5pm Europe on Thurs 28th April* - please reply to let me know if 
> you can make it.
>
> best wishes
> Andy
>

-- 
Robert Chalkley PhD
Adjunct Associate Professor
Genentech Hall, N474A
Tel: 415 476 5189
Fax: 415 502 1655

NIGMS Mass Spectrometry Facility
University of California San Francisco
600 16th Street,
Genentech Hall, suite N472A
San Francisco, CA 94143-2240