1. Refining Peptide Fragmentation Models for Improved Confidence in Sequence/Spectrum Matching Karl Clauser Broad Institute of MIT and Harvard Cambridge, MA
Strategies for Improving Reliability in Protein/Peptide Identification Workshop
National Cancer Institute's (NCI)
Clinical Proteomic Technologies for Cancer (CPTAC) Initiative
and the
National Institute of Standards and Technology (NIST)
Gaithersburg, MD
November 15, 2007
9/17/2018 7:50:32 AM

2. Continuum of MS/MS Interpretation Strategies
Unsequenced Genome
Unexpected PTM
Multiple Mutations
Validate DB match
Spectral Quality
Sequenced Genome
Expected PTM
Expected Chem Mods
Single Mutations
Repeat Sample Analysis
Related Spectra
Unexpected Partial Chem Mods
Overlapping Peptides
De Novo
Sequence Database Match
Spectral Library
Fragmentation Modeling
Peak Detection
Collection
Curation
Each Digestion Enzyme
Each Instrument Model
Each Collision Energy
SCORING
9/17/2018 7:50:32 AM

3. Are Spectral Similarity Scoring Functions Appropriate for Scoring Database Search Sequence/Spectrum Matches?
( S( Ai * Bi) )2
SAi 2 * S Bi 2
Difference of the 3 Scores
Extent of Emphasis on low abundance ions
Dot product
Sokolow, 1978
General Features of the Scores
Normalize Intensities & Number of Peaks
Score contribution for each ion scales with its intensity
Max Value is 1.0
Spectral
Contrast Angle
Wan, 2001
S( Ai * Bi)
( SAi 2 * SBi 2)1/2
Numerator
Bonus for Matched ions
Denominator
Penalty for Unmatched ions A Experimental
Penalty for Unmatched ions B Theoretically Possible
S( Ai * Bi ) 1/2
( SAi * SBi )1/2
Similarity
Zhang, 2004
The naïve fragment ion models in current database search programs tend to substantially over predict the number of fragment ions.
I.e. allow for all possible ion types with similar intensities
9/17/2018 7:50:32 AM

4. (R)L/Q|A|E|A|Q|Q|L|R(K)
Zhang
Kinetic Model
ZSS: 0.826
SM Core
Model
Experimental
9/17/2018 7:50:32 AM

5. (R)L/Q|A|E|A|Q|Q|L|R(K)
Zhang
Kinetic Model
ZSS: 0.826
SM Dominant Ion
Model v9
Experimental
9/17/2018 7:50:32 AM

6. Non-assignment Penalty
SpectrumMill Core Scoring of MS/MS Interpretations
(R)E F E|I|I|W|V T K(H)
9.24
78.1%
DEBS: -3
DFwdRev: 1.75
Score =
Assignment Bonus
(Ion Type Weighted) -
Non-assignment Penalty
(Intensity Weighted)
Peak Selection: De-Isotoping, S/N thresholding,
Parent - neutral removal, Charge assignment
Match to Database Candidate Sequences
SPI (%)
Scored Peak Intensity
9/17/2018 7:50:32 AM

7. Refining Peptide Fragmentation Models to Account for Dominant Peaks
Goal
Increase gap between score of correct and incorrect sequences for each spectrum
Increase score of correct sequence
Decrease score of incorrect sequence
Principles to Capture
Bonus - dominant peaks should score a bonus for ion types and AA adjacencies that should be dominant, i.e. at N-term of Pro.
Penalty - dominant peaks should be penalized if assigned to sites that should not be dominant.
Correct peak assignments should rarely be penalized (<1%)
Charge location - At a dominant site, formation of a b or y type ion depends primarily on the location of basic residues in the peptide, not the fragmentation site.
Scoring scheme should still work if an enzyme besides trypsin is used.
Do not try to predict which of b, y, b++, y++ will dominate
Frequency - dominant fragmentation does not always occur at particular AA adjacencies.
Instead there are tendencies, additional factors not yet understood.
How dominant - It is very difficult to predict how dominant a peak will be when it occurs at a cleavage that is expected to be dominant.
Do not try to predict the extent of dominance.
Scale a peak’s score with the extent of dominance that occurs.
9/17/2018 7:50:32 AM

8. Additional Intensity Scaled Score on Dominant Peaks
Bonus
b or y at expected
dominant cleavage
Penalty
not expected
dominant cleavage
M PM NM
n R 2 2 2
c R 0 0 2
n P 2 2 2
c DnE 1 2 2
nc HK 2 2 0
c ILV 1 1 0
AG, FG, QG, YG 1 0 0
n-1: HALIVFYW, !n: PG 1* 1* 0
b2/yn
bn-2 contains RHK 0 0 0
nterm E b-h2o 1 1 1
nterm Q b-nh
yn-2 DnEST 1 1 0
yn-2 QH 1 1 0
Dominant Ions
Expected by Peptide Sequence
*not counted as expected, but bonus if occurs
Proton Mobility
Mobile: zpre > #Arg + #Lys + #His
Partially mobile: zpre < #Arg + #Lys + #His and > #Arg
Non-mobile: zpre < #Arg
(amount to add to base score)
Bonus Score
1.5
0.75
0.25
0.125
0.0
Class 1
Class 2
Class 0
(amount to subtract from base score)
Penalty Score
1.5
0.25
0.0
Intensity
(Dominant Peak/ Tallest non-dominant peak)
Tallest
non-dominant =
peak
min(
seq: tallest not expected,
spectrum: top 5 ratio gap )
9/17/2018 7:50:32 AM

9. Frequency and Distribution of Dominant Ions v9
67%
72%
76%
5758
2974
177
Proton Mobility
Mobile: zpre > #Arg + #Lys + #His
Partially mobile: zpre < #Arg + #Lys + #His and > #Arg
Non-mobile: zpre < #Arg
Precursor z=2, 6699 spectra, <0.5% FPR
from a trypsin GeLC/MS/MS experiment
on an LTQ-FT
9/17/2018 7:50:32 AM

10. Frequency of AA Adjacency-Dependent Dominant Ions – v9, z=2
Mobile
Partially Mobile
4525 spectra
2061 spectra
# dominant ions
# total cleavages
>0.8
- (<3 obsv)
Non-mobile
114 spectra
9/17/2018 7:50:32 AM

11. Frequency of Position-Dependent Dominant Ions – v9, z=2
Proton Mobility
Mobile: zpre > #Arg + #Lys + #His
Partially mobile: zpre < #Arg + #Lys + #His and > #Arg
Non-mobile: zpre < #Arg
9/17/2018 7:50:32 AM

12. DIS performance Partially Mobile v9
P050531_PBD_WT
2061 spectra unique peptides
1477 spectra w/ dominant peaks
2060 spectra yield same match DIS v Orig
z=2 only, m/z tolerance = 0.05
DIS better
R1-R2 only
DIS better
both
DIS worse
both
DIS better score only
Score
Rank1 – Rank2
DIS worse
DIS worse
DIS better
DIS better
9/17/2018 7:50:32 AM

13. Spectral Similarity Comparison of Fragmentation Models
Zhang Kinetic Model
SM Core Model v9
9/17/2018 7:50:32 AM

14. Zhang Kinetic Model SM Dominant Ion Model v9 SM Core Model v9
9/17/2018 7:50:32 AM

15. Dominant Ions – Mobile b2/yn-2
(K)V/N|I|V|P V I\A K(A)
ZSS: 0.697
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

16. Dominant Ions – Mobile b2/yn-2
(K)A N|S/N/L/V L|Q|A|D\R(S)
ZSS: 0.534
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

17. Dominant Ions – PM nterm Q b-nh3
(K)Q D/A Q/S/L/H|G D|I|P Q\K(Q)
ZSS: 0.808
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

18. Dominant Ions – PM nterm E b-h2o
(K)E/G E/D/S/S/V/I/H|Y\D|D\K(A)
ZSS: 0.753
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

19. Dominant Ions – PM y++-h2o @ yn-2 DnEST
(K)V/A|E|I/E|H|A\E\K(E)
ZSS: 0.724
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

20. Dominant Ions – PM y++-nh3 @ yn-2 QH
(R)I/L|Q|E/H|E|Q\I\K(K)
ZSS: 0.779
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

21. Dominant Ions – PM bn-2 contains RHK
(R)R/Q V E|E\E|I\L|A|L\K(A)
ZSS: 0.799
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

22. Non-dominant Ions – Mobile
(R)T/V T V W E L I/S|S/E/Y|F|T|A|E\Q R(Q)
ZSS: 0.427
Zhang
Kinetic Model
Experimental
9/17/2018 7:50:32 AM

23. Next Steps Refine definitions of dominant adjacencies & positions
Add/combine factors to increase frequency of occurrence/expectation
Bonus Class Cleavage Matrix
Refine intensity ratio and dominant ion categories across charge states.
Verify suitability across multiple enzymes.
Verify suitability across multiple collision energies 28, 30, 35.
Test scoring system on phospho-peptides and other labeling chemistries that produce strong marker ions and/or bear charge.
Examine diminishing the relative score of minor ion types (a, b/y-h2o, b/y-nh3)
Minimize role of SPI (secondary score) in validation.
Evaluate/Integrate Zhang Fragment model
9/17/2018 7:50:32 AM

24. Acknowledgements Broad Institute Steve Carr Jake Jaffe MIT
Michael Yaffe
Majbrit Hjerrld
Drew Lowery
Agilent Technologies
Joe Roark
David Horn
Amgen
Zhongqi Zhang
9/17/2018 7:50:32 AM