EMBnet course Proteomics using Bioinformatics tools

EMBnet course
Proteomics using
Bioinformatics tools
MS identification tools
Patricia M. Palagi
PIG, SIB, Geneva
PMP

The data: list of m/z values
Peptide mass values and intensities
MS
840.6950 13.75
1676.9606 26.1
1498.8283 128.9
1045.564 845.2
2171.9670 2.56
861.1073 371.2
842.51458 53.7
1456.7274 12.9
863.268365 3.1
Parent mass value
fragment mass values
MS/MS
Parent mass charge
1163.7008 2
86.1105 220.1429
86.1738 13.7619
102.0752 4.3810
147.1329 57.3333
185.1851 649.0953
185.3589 5.3810
186.1876 81.4286
213.0791 1.4286
fragment intensities
PMP

PMP
The tools

One direct access to all- ExPASy
http://www.expasy.org/tools/
PMP

Automatic protein identification
- Peptide mass fingerprinting – PMF
- MS/MS sequence search
- MS/MS spectra library search
- MS/MS prospective analysis (tag,
open mod, de novo sequencing
PMP

Peptide mass fingerprinting = PMF
MS database matching
Protein(s)
Enzymatic
digestion
…MAIILAGGHSVRFGPKAF
AEVNGETFYSRVITLESTNM
FNEIIISTNAQLATQFKYPN
VVIDDENHNDKGPLAGIYTI
MKQHPEEELFFVVSVDTPM
ITGKAVSTLYQFLV …
Sequence
database entry
Peptides
In-silico
digestion
Mass spectra
- MAIILAGGHSVR
-FGPK
- AFAEVNGETFYSR
- VITLESTNMFNEIIISTNAQLATQFK
- YPNVVIDDENHNDK
…
Theoretical
proteolytic peptides
Peaklist
840.695086
1676.96063
1498.8283
1045.564
2171.967066
861.107346
842.51458
1456.727405
863.268365
Match
861.107346
838.695086
1676.96063
1498.8283
1045.564
2171.967066
842.51458
1457.827405
863.268453
Theoretical
peaklist
Result:
ranked list
of protein
candidates
PMP

Peptide mass fingerprinting
What you have:
- Set of peptide mass values
- Information about the protein: molecular weight, pI, species.
- Information about the experimental conditions: mass spectrometer
precision, calibration used, possibility of missed-cleavages, possible
modifications
- Biological characteristics: post-translational modifications, fragments
What will do the tool:
- Match between this information and a protein sequence database
What will you get:
- a list of probable identified proteins
PMP

What is the expected information in a
submission form?
• Place to upload a spectrum (many spectra)
• Description of the sample process used
– Chemical process such as alkylation/reduction,
– Cleavage properties (enzyme),
– Mass tolerance (m/z tolerance)
• Search space
– Sequence databank,
– taxonomy restriction
– Mw, pI restriction
• Scoring criteria and filters
PMP

One example of parameter
effects on the search
• Accepted mass tolerance
‣ due to imprecise measures and calibration problems
PMP Source: Introduction to proteomics: tools for the new biology. Daniel C. Liebler. Human Press. 2002

Summary of PMF tools
Tool
Aldente
Mascot
MS-Fit
ProFound
PepMAPPER
PeptideSearch
PepFrag
Source website
www.expasy.org/cgi-bin/aldente
www.matrixscience.com/
prospector.ucsf.edu/
prowl.rockefeller.edu/profound_bin/WebProFound.exe
wolf.bms.umist.ac.uk/mapper/
www.mann.emblheidelberg.de/GroupPages/PageLink/peptidesearchpage.html
prowl.rockefeller.edu/prowl/pepfragch.html
PMP
Non exhaustive list!

Scoring systems
• Essential for the identification! Gives a confidence value to
each matched protein
• Three types of scores
• Shared peaks count (SPC): simply counts the number of
matched mass values (peaks)
• Probabilistic scores: confidence value depends on
probabilistic models or statistic knowledge used during the match
(obtained from the databases)
• Statistic-learning: knowledge extraction from the influence of
different properties used to match the proteins (obtained from the
databases)
PMP

Mascot
http://www.matrixscience.com/
• Internet free version in the above website
(commercial versions available too)
•Choice of several databases.
• Considers multiple chemical modifications.
• 0 to 9 missed-cleavages.
• Score based on a combination of probabilistic
and statistic approaches (is based on Mowse
score).
• Considers Swiss-Prot annotations for Splice
Variants (in locally installed versions).
PMP

Mascot - principles
• Probability-based scoring
• Computes the probability P that a match is
random
• Significance threshold p< 0.05 (accepting that
the probability of the observed event occurring
by chance is less than 5%)
• The significance of that result depends on the
size of the database being searched.
• Mascot shades in green the insignificant hits
• Score: -10Log 10 (P)
PMP

Mascot
Input
PMP

Decoy
Output
Hints about the significance
of the score
PMP

Sequence coverage
Output
Peptides matched
Error
function
PMP

Aldente
• SwissProt/TrEMBL db, indexed masses (trypsine and many
others).
• Considers chemical modifications and user specified
modifications.
• Considers biological modifications (annotations SWISS-PROT).
• 0 or 1 missed-cleavages.
• Use of robust alignment method (Hough transform):
• Determines deviation function of spectrometer
• Resolves ambiguities
• Less sensitive to noise
PMP

Aldente – summary
Experimental masses / peaks
Spectrometer
calibration error
Spectrometer
internal error
• The Hough Transform estimates from
the experimental data the deviation
function of the mass spectrometer (the
calibration error function).
Theoretical masses / peptides
• The program optimizes the set of
best matches, excluding noise and
outliers, to find the best alignment.
PMP

PMP
Aldente - Input

PMP
Aldente - Input

PMP
Aldente - Input

PMP
Aldente - Input

PMP
Aldente - Input

Output
Hints about
the
significance of
the score
PMP

PMP
Information from Swiss-Prot
annotation. Processed protein (signal
peptide is cleaved).

PMP
BioGraph

What is the expected information
in an identification result?
• A summary of the search parameters
• A list of potentially identified proteins (AC numbers) with
scores and other evidences
• A detailed list of potentially identified peptides (associated or
not to the potentially identified proteins) with scores
• Possibilities to validate/invalidate the provided results (info on
the data processing, on the statistics, links to external
resources, etc.)
• Possibilities to export the (validated) data in various formats
PMP

Protein characterization with PMF
data
1 protein entry
does not represent
1 unique molecule
- Exact primary structure
- Splicing variants
- Sequence conflicts
-PTMs
Characterization tools at ExPASy using peptide mass fingerprinting data
http://www.expasy.org/tools/
FindMod
GlycoMod
FindPept
Prediction tools
• PTMs and AA substitutions
• Oligosaccharide structures
• Unspecific cleavages
PMP

PMP
SWISS-PROT feature table:
active protein is more than just translation of
gene sequence (example: P20366)

PMP
Detection of PTMs in MS
769.8
893.4
1326.7
1501.9
2100.6
1056.1
624.3
624.3
769.8
893.4
994.5
994.5
1056.1
1326.7
1501.9
1759.8
1759.8 1923.4
1923.4
2100.6
Unmodified
tryptic
masses
600 2200
Δ m/z
=> PTM
769.8
769.8
893.4
893.4
1056.1
1070.1
1326.7
1326.7
1501.9
1501.9
2100.6
2100.6
624.3
624.3
994.5
994.5
1759.8
1759.8 1923.4 1923.4
Tryptic
masses of
a modified
protein
600 2200

PMP
FindMod http://www.expasy.org/tools/findmod/
AA modifications
DB entry
experimental
options
experimental
masses

FindMod Output
}unmodified peptides,
modified peptides
known in SWISS-PROT
and chemically modified
peptides
}
putatively modified
peptides predicted
by mass differences
+ putative AA substitutions
PMP

Modification rules can be defined from
SWISS-PROT, PROSITE and the literature
some examples:
modification amino acid rule exceptions
farnesylation Cys -
palmitoylation Cys Ser, Thr
O-GlcNAc Ser, Thr Asn
amidation
pyrrolidone carboxylic acid Gln (N-term) -
Xaa (C-term) where Gly followed Xaa
phosphorylation in eukaryotes: Ser, Thr, Asp, His, Tyr -
in prokaryotes: Ser, Thr, Asp, His, Cys -
PMP
sulfatation in eukaryotes
Tyr, PROSITE PDOC00003

FindMod Output - Application of Rules
- potentially modified peptides that agree with rules are listed
- amino acids that potentially carry modifications are shown
- peptides potentially modified only by mass difference
PMP
- predictions can be tested by MS-MS peptide fragmentation

FindPept
http://www.expasy.org/tools/findpept.html
• From MS (peptide mass fingerprint) data -
detection of :
– Matching peptides for unspecific cleavage
– Masses resulting from possible contaminants
– Matching peptides for specific cleavage (16 different
enzymes)
– Peptides resulting from protease autolysis
PMP

PMP
FindPep

PMP
FindPep

PMP
FindPep

PMP
FindPep

MS/MS based identification tools
• Tag search- Tools that search peptides based on a MS/MS
Sequence Tag
– MS-Tag and MS-Seq, PeptideSearch
• Ion search or PFF - Tools that match MS/MS experimental
spectra with “theoretical spectra” obtained via in-silico
fragmentation of peptides generated from a sequence
database
– Phenyx, Mascot, Sequest, X!Tandem, OMSSA, ProID, …
• de novo sequencing - Tools that directly interpret MS/MS
spectra and try to deduce a sequence
– Convolution/alignment (PEDENTA)
– De-novo sequencing followed by sequence matching
(Peaks, Lutefisk, Sherenga, PeptideSearch)
– Guided Sequencing (Popitam)
In all cases, the output is a peptide structure per MS/MS spectrum
PMP

Peptide fragmentation
fingerprinting = PFF = ion search
MS/MS database matching
Protein(s)
…MAIILAGGHSVRFGPKAF
AEVNGETFYSRVITLESTNM
FNEIIISTNAQLATQFKYPN
VVIDDENHNDKGPLAGIYTI
MKQHPEEELFFVVSVDTPM
ITGKAVSTLYQFLV …
PMP
Enzymatic
digestion
Sequence
database entry
In-silico
digestion
Peptides
- MAIILAGGHSVR
-FGPK
- AFAEVNGETFYSR
- VITLESTNMFNEIIIK
- YPNVVIDDENNDK
…
Theoretical
proteolytic peptides
MS/MS spectra
of peptides
In-silico
fragmentation
-MAIILAG
-MAIILA
-MAIIL
-MAII
-MAI
-M
-M
-AIILAG
Theoretical
fragmented
peptides
Ions peaklists
340.695086
676.96063
498.8283
545.564
1171.967066
261.107346
342.51458
456.727405
363.268365
Match
361.107346
338.695086
676.96063
498.8283
1045.564
1171.967066
342.51458
457.827405
263.268453
Theoretical
peaklist
Result:
ranked list
of peptide
and
protein
candidates

Ion-types
offset
-28
-45
-46
0
-17
-18
+17
+28
+ 2
-15
-16
-15
P' nterm
P' cterm
It is very important to know the ionic series
produced by a spectrometer, otherwise
potential matches will be missed.
In the other hand, if an ion-type not present
in the original spectrum is taken into account,
it will contribute to get false positive
matches.
[N] is the mass of the N-term group
[M] is the mass of the sum of the neutral amino acid residue
masses
PMP

Some PFF tools
Same principle of a PMF, but using MS/MS spectra
Software
InsPecT
Mascot
MS-Tag and MS-Seq
PepFrag
Phenyx
Popitam
ProID (download)
Sequest*
Sonar
SpectrumMill*
VEMS
X!Tandem (download)
Source website
peptide.ucsd.edu/inspect.py
www.matrixscience.com/search_form_select.html
prospector.ucsf.edu
prowl.rockefeller.edu/prowl/pepfragch.html
phenyx.vital-it.ch
www.expasy.org/tools/popitam
sashimi.sourceforge.net/software_mi.html
fields.scripps.edu/sequest/index.html
65.219.84.5/service/prowl/sonar.html
www.home.agilent.com
www.bio.aau.dk/en/biotechnology/vems.htm
www.thegpm.org/TANDEM
PMP
*Commercialized
Non exhaustive list!

PMP

PMP
Phenyx

Submission
The Phenyx Web Interface:
One result, multiple views
Desktop
Results
views
Results comparison
Management console
PMP P.A. Binz
Excel, xml and text exports

The Proteins overview
List of identified
proteins
Protein group
description
Corresponding list of
identified peptides
PMP P.A. Binz

The Proteins overview
Hints about the
significance of
the score
PMP

Hints about the
significance of
the score
PMP
Better when high intensity peaks are matched and ion series are
extended, without too many and too big holes

The scoring system in Phenyx
• The score is the sum of up to 12 basic scores such as:
– presence of a, b, y, y++, B-H 2
O…; co-occurrence of ion series
(using HMMs), peak intensities, residue modifications (PTM or
chemical), …
• True probabilistic approach for each peptide match
(likelihood of being correct)
Search in a query database
log --------------------------------
(likelihood of being random)
Search in a randomized set
of peptides
• Function of instruments and molecular types
– Esquire 3000+, LCQ; iTRAQ vs. unmodified peptides
• Scores are normalised into z-scores
PMP

X!Tandem
PMP
www.thegpm.org

X!Tandem - output
1
2
3
PMP

The two-rounds search
Mascot, Phenyx and X!Tandem
The identification process may be launched in 2-rounds
• Each round is defined with a set of search criteria
– First round searches the selected database(s) with
stringent parameters,
– Second round searches the proteins that have
passed the first round (relaxed parameters):
⇒Accelerate the job when looking for many variable modifications,
or unspecific cleavages
⇒Appropriate when the first round defines stringent criteria to
capture a protein ID, and the second round looks for looser
peptide identifications
PMP

Example 2 nd round
1rnd,
Only 3 fixed mods
131 valid,
75% cov.
2rnd,
Add variable mods
205 valid,
84% cov.
2rnd,
With all mods
And half cleaved
348 valid,
90% cov.
PMP

Source of errors in assigning
peptides
• Scores not adapted
• Parameters are too stringent or too loose
• Low MS/MS spectrum quality (many noise peaks, low
signal to noise ratio, missing fragment ions, contaminants)
• Homologous proteins
• Incorrectly assigned charge state
• Pre-selection of the 2 nd isotope (the parent mass is shifted
of 1 Da. A solution is to take the parent mass tol. larger, but
may drawn the good peptide too)…
• Novel peptide or variant
PMP

Hints to know when the
PMP
With MS
identification is correct
• Good sequence coverage: the larger the sub-sequences and the
higher the sequence coverage value, the better
• Consider the length of the protein versus the number of
matched theoretical peptides
•Better when high intensity peaks have been used in the identification
•Scores: the higher, the better. The furthest from the 2nd hit the
better
• Filter on the correct species if you know it (reduces the search
space, time, and errors)
• Better when the errors are more or less constants among all
peptides found.
•If you have time, try many tools and compare the results

With MS/MS
Hints to know when the
identification is correct
• The higher the number of peptides identified per protein, the better
• Sequence coverage: the larger the sub-sequences and the higher
the sequence coverage value, the better
•Depends on the sample complexity and experiment workflow
• Scores: the higher, the better.
• Filter on the correct species if you know it (reduces the search
space, time, and errors)
• Better when high intensity peaks are matched and ion series are
extended, without too many and too big holes.
• Better when the errors are more or less constants among all ions.
• If you have time, try many tools and compare the results
PMP

E-values
• For a given score S, it indicates the number of
matches that are expected to occur by chance
in a database with a score at least equal to S.
• The e-value takes into account the size of the
database that was searched. As a consequence
it has a maximum of the number of sequences
in the database.
• The lower the e-value, the more significant the
score is.
PMP
• An e-value depends on the calculation of the p-
value.

p-value
• A p-value describes the probability, which
assesses the chance of validly rejecting the null
hypothesis. If the p-value is 10 -5 then the
rejection of the null hypothesis is due to chance
with a probability of 10 -5 .
• A p-value has a maximum of 1.0.
• The larger the search space, the higher the p-
value since the chance of a peptide being a
random match increases.
• The lower the p-Value, the more significant is the
match.
PMP Source: Lisacek, Practical Proteomics, 2006 Sep;6 Suppl 2:22-32

Z-score
• Z-score is a dimensionless quantity
derived by subtracting the population
mean from an individual (raw) score and
then dividing the difference by the
population standard deviation.
Z − score =
• The z score reveals how many units of
the standard deviation a case is above or
below the mean.
x
− μ
σ
PMP Source: wikipedia

So what?
• For small (significant) p-values, p and e are
approximately equal, so the choice of one or the
other is often equivalent. It is therefore
reasonable to assimilate low p-values in Phenyx
to e-values. X!Tandem simply switches e-values
to log values to remove the powers of 10
• For a single search (or set of sampled peptides),
you can compare z-scores. However, when two or
more searches are performed on different size
spaces, you first need to look at the p-values
before comparing z-scores.
PMP Source: Lisacek, Practical Proteomics, 2006 Sep;6 Suppl 2:22-32