Download A toolbox for validation of mass spectrometry peptides identification

BIOINFORMATICS APPLICATIONS NOTE
Vol. 25 no. 15 2009, pages 1980–1981
doi:10.1093/bioinformatics/btp301
Gene expression
A toolbox for validation of mass spectrometry peptides
identification and generation of database: IRMa
Véronique Dupierris1 , Christophe Masselon2 , Magali Court2 , Sylvie Kieffer-Jaquinod2 and
Christophe Bruley2,∗
Rhône-Alpes Futur, 89 rue Bellecombe, 69003 Lyon and 2 CEA, DSV, iRTSV, Laboratoire d’Etude de la
Dynamique des Protéomes, INSERM, U880, Université Joseph Fourier, Grenoble F-38054, France
1 Fondation
Received on December 5, 2008; revised and accepted on April 30, 2009
Advance Access publication May 6, 2009
Associate Editor: John Quackenbush
ABSTRACT
Summary: The IRMa toolbox provides an interactive application
to assist in the validation of Mascot search results. It allows
automatic filtering of Mascot identification results as well as manual
confirmation or rejection of individual PSM (a match between a
fragmentation mass spectrum and a peptide). Dynamic grouping and
coherence of information are maintained by the software in real time.
Validated results can be exported under various forms, including an
identification database (MSIdb). This allows biologists to compile
search results from a whole study in a unique repository in order to
provide a summarized view of their project. IRMa also features a fully
automated version that can be used in a high-throughput pipeline.
Given filter parameters, it can delete hits with no significant PSM,
regroup hits identified by the same peptide(s) and export the result
to the specified format without user intervention.
Availability: http://biodev.extra.cea.fr/docs/irma (java 1.5 or higher
needed)
Contact: [email protected]
1
INTRODUCTION
The Mascot search engine uses mass spectrometry data to identify
proteins by matching experimental and theoretical peptide mass
spectra (Aebersold and Mann, 2003). However, it is necessary to
eliminate false matches (Baldwin, 2004; Nesvizhskii and Aebersold,
2004) in order to avoid incorrect protein identification.
The Mascot html output is not designed for this kind of
validation, and exporting the data to Excel for subsequent
validation results in loss of data consistency. Moreover, validation
being the first step in data interpretation, subsequent data mining
often requires compilation of numerous search results. IRMa is
a validation toolbox for Mascot identification results that allow
users to validate and export the identified proteins and their PSM to
various formats. Its main originality is to filter matches rather than
identified proteins.
2
VALIDATING HITS VERSUS PSM
For each submitted spectrum (query), Mascot attempts to match
up to 10 peptides from a sequence database. A score, based on the
∗ To
whom correspondence should be addressed.
1980
absolute probability that the observed match is a random event, is
assigned to each peptide spectrum matches (PSMs). For each query,
PSMs are ranked according to their score. Mascot then groups
PSMs into protein hits. A hit contains not only all proteins covered
by the same set of PSMs but also all proteins covered by a subset
of these PSMs. A hit score, based on its matching PSM scores, is
assigned to each protein.
While existing tools (Bouyssié et al., 2007) validate whole hits,
selecting confident PSMs and rejecting dubious ones undoubtedly
allow more accurate identification. When validating whole hits,
false PSMs can still be considered in the result. On the other
hand, validation at the PSM level affects the grouping consistency
(and hits properties such as score and coverage): rejecting PSM
without grouping update leads to redundancy of the protein hits’
list. IRMa allows validation at the PSM level, while circumventing
the associated difficulties.
3
FEATURES
IRMa uses the Mascot Parser distributed free of charge by Matrix
Science to build an identification result from the native ‘.dat’ files
generated by the Mascot server. The parser can be invoked with
different report parameters, such as the number of hits or significance
thresholds. At this stage, the same information and grouping shown
in the Mascot result page are available but displayed in a structured
and accessible layout by IRMa (see Fig. 1). PSMs relevant for
protein identifications are flagged as significant. A PSM having the
same sequence as a significant one but associated to another query
and with a lower score are flagged as duplicated.
For validation purposes, in addition to the significant and
duplicated categories, IRMa introduces a third category in which
ambiguous PSMs are classified. This third category regroups
PSMs rejected by filtering and that do not contribute to protein
identification. While PSMs’ classification into these categories
can be done manually, several filters applied on the Mascot
parser results are proposed. They provide rules to assign PSMs
to significant or ambiguous categories and can be based on PSM
ranks, scores, on mandatory post-translational modifications or
on PSMs properties expression. This automatic classification can
always be adjusted; the user can restore PSMs from the ambiguous
group to the significant group or vice versa. Nevertheless, IRMa
will ensure that the exact same sequence cannot appear twice
© The Author 2009. Published by Oxford University Press. All rights reserved. For Permissions, please email: [email protected]
[18:08 26/6/2009 Bioinformatics-btp301.tex]
Page: 1980
1980–1981
IRMa
4
Fig. 1. IRMa snapshot. User can browse the submitted spectra (queries)
(1) or the identified protein hits and PSMs (2). The main component of the
interface displays query statistics or, as shown here, hit information. In this
case, the upper part (3) shows the master protein description and properties
together with the lists of same-set and sub-set proteins. PSMs classified in
significant, ambiguous and duplicated categories are displayed for the current
hit (4).
Biological studies often require the comparison of numerous
samples. Furthermore, each sample can be fractionated before
analysis to generate several aliquots, which in turn result in as many
Mascot searches. One of the major challenges facing the biologist
resides in the high redundancy in identification lists. Compiling
multiple search results through multiple Excel files or html results
pages is impractical. In order to overcome this difficulty, the creation
of a unique database, containing the identification results of a
whole project, allows the user to access information on all analyzed
samples at once.
IRMa was used in our laboratory to compile data from a largescale proteomics study of A. thaliana chloroplasts. The MSIdb
was compiled in a fully unattended mode after providing the
software with filter parameters and an output path. The database
was populated using search results from ∼585 000 Mascot queries
generated from ∼500 LC–MS/MS analyses, with minimal user
intervention.
5
in the significant category. Dynamic classification has an impact
on protein hits: when a PSM is declared ambiguous, it becomes
irrelevant to proteins identification. IRMa takes into account these
changes to ensure consistency of information such as protein
coverage and identification score and dynamically proposes new
protein hit sets and sub-sets based on significant PSMs. For
instance, two proteins of a same hit that initially shared only
a sub-set of PSMs can become indistinguishable (all significant
PSMs are shared) if specific PSMs are judged ambiguous. Another
possibility is the merge of two distinct hits into a unique one
if discriminating PSMs are flagged as ambiguous. For example,
given the following two hits, ACTG_HUMAN {pep_63, pep_133,
pep_167, pep_254, pep_311} and Q53G99_HUMAN {pep_63,
pep_133,pep_139, pep_254, pep_311}, if PSM pep_139 is declared
ambiguous, IRMa will propose grouping Q53G99_HUMAN and
ACTG_HUMAN as the same hit.
The user can validate new grouping proposed by the application
by merging hits that become sub-sets or same sets of another and
by deleting hits that no longer contain any significant PSM. This
implies that hits are reordered according to their scores.
The validated result can finally be exported under various forms:
to the widely used Microsoft Excel format; to a PDF format
including graphical representation of annotated spectra1 or to
MSIdb2 . The information recorded in all exports can be customized.
For instance, in Microsoft Excel format, user can choose which
worksheets (protein summary, similar proteins, peptide list, etc.) and
which columns will be present. Information on Mascot search
parameters or automatic filtering is also reported in the output file.
The MSIdb stores this information as well as that on grouping
and matches between proteins and PSMs. Exporting to MSIdb is
especially suitable for large-scale studies.
1
2
LARGE-SCALE STUDIES
PERSPECTIVES
IRMa allows generation of an MSIdb for a given project. To
enhance data mining of comprehensive identification catalog, it
would therefore be helpful to integrate identification databases
with information from other data sources (e.g. web-based public
repositories). We are currently working on a tool to achieve
this aim and to further investigate identified proteins. Grouping
identifications at the whole database level to remove redundancy
among samples or comparing multiple experiments are some of the
available operations.
ACKNOWLEDGEMENTS
We thank Matrix Science technical support team for their timely
answers to our requests, the IRMa workgroup for their help in
application specification and design, and all EDyP Laboratory team
members for testing and feedback.
Funding: CEA, INSERM and the Rhône-Alpes Génopôle.
Conflict of Interest: none declared.
REFERENCES
Aebersold,R. and Mann,M. (2003) Mass spectrometry-based proteomics. Nature, 422,
198–207.
Baldwin,M. (2004) Protein identification by mass spectrometry: issues to be considered.
Mol. Cell. Proteomics 3, 1–9.
Bouyssié,D. et al. (2007) MFPaQ, a new software to parse, validate, and quantify
proteomic data generated by ICAT and SILAC mass spectrometric analyses:
application to the proteomic study of membrane proteins from primary human
endothelial cells. Mol. Cell. Proteomics 6, 1621–1637.
Nesvizhskii,A.I. and Aebersold,R. (2004) Analysis, statistical validation and
dissemination of large-scale proteomics datasets generated by tandem MS. Drug
Discov. Today, 9, 173–181.
MCP guideline compliant - for single peptide hits and PTM peptides hits.
For database structure see supplementary material.
1981
[18:08 26/6/2009 Bioinformatics-btp301.tex]
Page: 1981
1980–1981