Protein Sequence Databases, Peptides to Proteins, and Statistical Significance

1
Protein Sequence Databases, Peptides to Proteins,
and Statistical Significance

• Nathan Edwards
• Department of Biochemistry and Mol. Cell.
  Biology
• Georgetown University Medical Center

2
Protein Sequence Databases

• Link between mass spectra and proteins
• A proteins amino-acid sequence provides a basis
  for interpreting
• Enzymatic digestion
• Separation protocols
• Fragmentation
• Peptide ion masses
• We must interpret database information as
  carefully as mass spectra.

3
More than sequence

• Protein sequence databases provide much more than
  sequence
• Names
• Descriptions
• Facts
• Predictions
• Links to other information sources
• Protein databases provide a link to the current
  state of our understanding about a protein.

4
Much more than sequence

• Names
• Accession, Name, Description
• Biological Source
• Organism, Source, Taxonomy
• Literature
• Function
• Biological process, molecular function, cellular
  component
• Known and predicted
• Features
• Polymorphism, Isoforms, PTMs, Domains
• Derived Data
• Molecular weight, pI

5
Database types
Curated Swiss-Prot UniProt RefSeq NP Translated TrEMBL RefSeq XP, ZP
Omnibus NCBIs nr MSDB IPI Other PDB HPRD EST Genomic
6
SwissProt

• From ExPASy
• Expert Protein Analysis System
• Swiss Institute of Bioinformatics
• 515,000 protein sequence entries
• 12,000 species represented
• 20,000 Human proteins
• Highly curated
• Minimal redundancy
• Part of UniProt Consortium

7
TrEMBL

• Translated EMBL nucleotide sequences
• European Molecular Biology Laboratory
• European Bioinformatics Institute (EBI)
• Computer annotated
• Only sequences absent from SwissProt
• 10.5 M protein sequence entries
• 230,000 species
• 75,000 Human proteins
• Part of UniProt Consortium

8
UniProt

• Universal Protein Resource
• Combination of sequences from
• Swiss-Prot
• TrEMBL
• Mixture of highly curated/reviewed (SwissProt)
  and computer annotation (TrEMBL)
• Similar sequence clusters are available
• 50, 90, 100 sequence similarity

9
RefSeq

• Reference Sequence
• From NCBI (National Center for Biotechnology
  Information), NLM, NIH
• Integrated genomic, transcript, and protein
  sequences.
• Varying levels of curation
• Reviewed, Validated, , Predicted,
• 9.7 M protein sequence entries
• 209,000 reviewed, 90,000 validated
• 39,000 Human proteins

10
RefSeq

• Particular focus on major research organisms
• Tightly integrated with genome projects.
• Curated entries NP accessions
• Predicted entries XP accessions
• Others YP, ZP, AP

11
IPI

• International Protein Index
• From EBI
• For a specific species, combines
• UniProt, RefSeq, Ensembl
• Species specific databases HInv-DB, VEGA, TAIR
• 87,000 (from 307,000 ) human protein sequence
  entries
• Human, mouse, rat, zebra fish, arabidopsis,
  chicken, cow
• Slated for closure November 2010, but still
  going

12
MSDB

• From the Imperial College (London)
• Combines
• PIR, TrEMBL, GenBank, SwissProt
• Distributed with Mascot
• so well integrated with Mascot
• 3.2M protein sequence entries
• Similar sequences suppressed
• 100 sequence similarity
• Not updated since September 2006 (obsolete)

13
NCBIs nr

• non-redundant
• Contains
• GenBank CDS translations
• RefSeq Proteins
• Protein Data Bank (PDB)
• SwissProt, TrEMBL, PIR
• Others
• Similar sequences suppressed
• 100 sequence similarity
• 10.5 M protein sequence entries

14
Human Sequences

• Number of Human genes is believed to be between
  20,000 and 25,000

SwissProt 20,000
RefSeq 39,000
TrEMBL 75,000
IPI-HUMAN 87,000
MSDB 130,000
nr 230,000
15
DNA to Protein Sequence
Derived from http//online.itp.ucsb.edu/online/inf
obio01/burge
16
UCSC Genome Browser

• Shows many sources of protein sequence evidence
  in a unified display

17
Accessions

• Permanent labels
• Short, machine readable
• Enable precise communication
• Typos render them unusable!
• Each database uses a different format
• Swiss-Prot P17947
• Ensembl ENSG00000066336
• PIR S60367 S60367
• GO GO0003700

18
Names / IDs

• Compact mnemonic labels
• Not guaranteed permanent
• Require careful curation
• Conceptual objects
• ALBU_HUMAN
• Serum Albumin
• RT30_HUMAN
• Mitochondrial 28S ribosomal protein S30
• CP3A7_HUMAN
• Cytochrome P450 3A7

19
Description / Name

• Free text description
• Human readable
• Space limited
• Hard for computers to interpret!
• No standard nomenclature or format
• Often abused.
• COX7R_HUMAN
• Cytochrome c oxidase subunit VIIa-related
  protein, mitochondrial Precursor

20
FASTA Format

• gt
• Accession number
• No uniform format
• Multiple accessions separated by
• One line of description
• Usually pretty cryptic
• Organism of sequence?
• No uniform format
• Official latin name not necessarily used
• Amino-acid sequence in single-letter code
• Usually spread over multiple lines.

21
FASTA Format
22
Organism / Species / Taxonomy

• The proteins organism
• or the source of the biological sample
• The most reliable sequence annotation available
• Useful only to the extent that it is correct
• NCBIs taxonomy is widely used
• Provides a standard of sorts Heirachical
• Other databases dont necessarily keep up
• Organism specific sequence databases starting to
  become available.

23
Organism / Species / Taxonomy

• Buffalo rat
• Gunn rats
• Norway rat
• Rattus PC12 clone IS
• Rattus norvegicus
• Rattus norvegicus8
• Rattus norwegicus
• Rattus rattiscus
• Rattus sp.

• Rattus sp. strain Wistar
• Sprague-Dawley rat
• Wistar rats
• brown rat
• laboratory rat
• rat
• rats
• zitter rats

24
Controlled Vocabulary

• Middle ground between computers and people
• Provides precision for concepts
• Searching, sorting, browsing
• Concept relationships
• Vocabulary / Ontology must be established
• Human curation
• Link between concept and object
• Manually curated
• Automatic / Predicted

25
Gene Ontology

• Hierarchical
• Molecular function
• Biological process
• Cellular component
• Describes the vocabulary only!
• Protein families provide GO association
• Not necessarily any appropriate GO category.
• Not necessarily in all three hierarchies.
• Sometimes general categories are used because
  none of the specific categories are correct.

26
Gene Ontology
27
Protein Families

• Similar sequence implies similar function
• Similar structure implies similar function
• Common domains imply similar function
• Bootstrap up from small sets of proteins/domains
  with well understood characteristics
• Usually a hybrid manual / automatic approach

28
Protein Families
29
Protein Families
30
Sequence Variants

• Protein sequence can vary due to
• Polymorphism
• Alternative splicing
• Post-translational modification
• Sequence databases typically do not capture all
  versions of a proteins sequence

31
Swiss-Prot Variant Annotations
32
Swiss-Prot Variant Annotations
33
Omnibus Database Redundancy Elimination

• Source databases often contain the same sequences
  with different descriptions
• Omnibus databases keep one copy of the sequence,
  and
• An arbitrary description, or
• All descriptions, or
• Particular description, based on source
  preference
• Good definitions can be lost, including taxonomy

34
Description Elimination

• gi12053249embCAB66806.1 hypothetical protein
  Homo sapiens
• gi46255828gbAAH68998.1 COMMD4 protein Homo
  sapiens
• gi42632621gbAAS22242.1 COMMD4 Homo
  sapiens
• gi21361661refNP_060298.2 COMM domain
  containing 4 Homo sapiens
• gi51316094spQ9H0A8COM4_HUMAN COMM domain
  containing protein 4
• gi49065330embCAG38483.1 COMMD4 Homo
  sapiens

35
Peptides to Proteins
Nesvizhskii et al., Anal. Chem. 2003
36
Peptides to Proteins
37
Peptides to Proteins

• A peptide sequence may occur in many different
  protein sequences
• Variants, paralogues, protein families
• Separation, digestion and ionization is not well
  understood
• Proteins in sequence database are extremely
  non-random, and very dependent

38
Indistinguishable Protein Sequences
Nesvizhskii, Aebersold, Mol Cell Proteomics, 2005
39
Indistinguishable Protein Sequences
Nesvizhskii, Aebersold, Mol Cell Proteomics, 2005
40
Protein Families
Nesvizhskii, Aebersold, Mol Cell Proteomics, 2005
41
Protein Grouping Scenarios

• Parsimony
• Minimum of proteins
• Weighted
• Choose proteinswith the most confident
  peptides(ProteinProphet)
• Show all
• Mark repeated peptides
• Often no (ideal) resolution is possible!

Nesvizhskii, Aebersold, Mol Cell Proteomics, 2005
42
High Quality Peptide Identification E-value lt
10-8
43
Moderate quality peptide identification E-value
lt 10-3
44
Peptide Identification

• Peptide fragmentation by CID is poorly understood
• MS/MS spectra represent incomplete information
  about amino-acid sequence
• I/L, K/Q, GG/N,
• Correct identifications dont come with a
  certificate!

45
Peptide Identification

• High-throughput workflows demand we analyze all
  spectra, all the time.
• Spectra may not contain enough information to be
  interpreted correctly
• bad static on a cell phone
• Peptides may not match our assumptions
• its all Greek to me
• Dont know is an acceptable answer!

46
What scores do wrong peptides get?

• Generate random peptide sequences
• Real looking fragment masses
• Empirical distribution
• Require similar precursor mass
• Arbitrary score function can model anything we
  like!

47
Random Peptide Scores
Fenyo Beavis, Anal. Chem., 2003
48
Random Peptide Scores
Fenyo Beavis, Anal. Chem., 2003
49
Random Peptide Scores

• Truly random peptides dont look much like real
  peptides
• Just use peptides from the sequence database!
• Assumptions
• IID sampling of score values per spectra
• Caveats
• Correct peptide (non-random) may be included
• Peptides are not independent

50
Extrapolating from the Empirical Distribution

• Often, the empirical shape is consistent with a
  theoretical model

Fenyo Beavis, Anal. Chem., 2003
Geer et al., J. Proteome Research, 2004
51
E-values vs p-values

• Need to adjust for the size of the sequence
  database
• Best false/random score goes up with number of
  trials
• E-value makes this adjustment
• Expected number of incorrect peptides (with this
  score) from this sequence database.
• E-value Trials p-value (to 1st approx.)

52
False Discovery Rate

• Which peptide IDs to accept?
• E-value only provides a per-spectrum statistic
• With enough spectra, even these can be
  misleading!
• Decide which spectra (w/ scores) will be
  accepted
• SEQUEST Xcorr, E-value, Score, etc., plus...
• Threshold on identification criteria
• Control the proportion of incorrect
  identifications in the result for entire dataset

53
Distribution of scores over all spectra
Brian Searle, Proteome Software
54
Distribution of scores over all spectra
False
True
Brian Searle, Proteome Software
55
False Discovery Rate

• FDRscore x false ids with score x
• all ids with score
  x
• Need to estimate numerator!
• Assumes the false (and true) scores, sampled over
  spectra, are IID
• Not true for some peptide-spectrum scores
• (Mostly) true for E-values
• Can compute the false ids using a decoy search
  

56
Peptide Prophet
Keller et al., Anal. Chem. 2002
Distribution of spectral scores in the results
57
Decoy searches

• Shuffle or reverse sequence database
• Same size as original
• Known false identifications
• Estimate False distribution
• Alternatively, merge targetdecoy results
• Competition between target and decoy scores
• Assume false target and false decoys each win
  half the time
• FDRscore x 2 decoy ids with score x
• target ids with
  score x

58
Summary

• Protein sequence databases have varying
  characteristics, choose wisely!
• Inferring proteins from peptides can be (very)
  tricky!
• Statistical significance can help control the
  proportion of errors in the (peptide-level)
  results.