�ݺ�ߣ

http://www.bits.vib.be/training

sequence databases

lennart martens
lennart.martens@ugent.be

Computational Omics and Systems Biology Group
Department of Medical Protein Research, VIB
Department of Biochemistry, Ghent University
Lennart Martens BITS MS Data Processing – Sequence Databases
Ghent, Belgium
lennart.m artens@ugent.be UGent, Gent, Belgium – 16 December 2011

PEPTIDES AND REDUNDANCY
IN SEQUENCE DATABASES


Peptide-level sequence redundancy

>Protein 1 >Protein 1 (1-6)
LENNARTMARTENS LENNAR
>Protein 2 >Protein 1 (7-10)
LENNARTMARTENT TMAR
>Protein 1 (11-14)
TENS
=
non-redundant protein DB
>Protein 2 (1-6)
LENNAR
=
≠ >Protein 2 (7-10)
TMAR
>Protein 2 (11-14)
non-redundant peptide DB
TENT

Database content: all peptide sequences in the database
Database inform ation: number of unique peptide sequences
database information
Database inform ation ratio:
database content

Information ratios for common databases
12,000,000 100%
93%
ratio Content information
10,307,319 90%

10,000,000 Tryptic cleavage, 1 allowed missed cleavage,
Mass limits from 600 to 4000 Da. 80%

70%
8,000,000

60%

6,000,000 50%
45%
41% 42%
40%

4,000,000 4,472,356
3,491,778
3,186,806 30%
23%

20%
2,000,000 1,584,806
2,394,844
1,877,500
1,559,685 10%
1,466,927
1,309,625

0 0%
UniProtKB/SwissProt UniProtKB/TrEMBL Ensembl human IPI human NCBI nr human
human human


ENRICHING SEQUENCE DATABASES


The influence of the sequence database

N C
In vivo processing Search
ID miss
base

N C
+
Enzymatic digest and subsequent
NH2-terminal peptide isolation

Not in the sequence database!


An example

Mitochondrial Isovaleryl-coA Dehydrogenase

MATATRLLGWRVASWRLRPPLAGFVS
N -term inal transit peptide (1-29)
30 47
QRAHSLLPVDDAINGLSEEQRQLRE…
I sovaleryl-CoA dehydrogenase (30 – 423)

…LDGIQCFGGNGYINDFPMGRFLRDA
423
KLYEIGAGTSEVRRLVIGRAFNADFH


Extending the information content

AHSLLPVDDAINGLSEEQR AHSLLPVDDAINGLSEEQR
HSLLPVDDAINGLSEEQR
SLLPVDDAINGLSEEQR
LLPVDDAINGLSEEQR
LPVDDAINGLSEEQR
PVDDAINGLSEEQR
VDDAINGLSEEQR
……

Revised search
Search
ID miss base
base
ID


Another example: in vivo protein cleavage

NH 2 COOH
R R
R D R

Caspase cleavage of this protein
(for 50%)

NH 2 COOH
R R
R D R

NH 2 COOH NH 2 COOH
R R
RD R

NH2-terminal peptide isolation

COOH COOH
NH 2 NH 2
R R

NOT IN DB!


Solving the issue: bifunctional enzymes

COOH
NH 2
R

result of in vivo result of in vitro
protease trypsin

Creation of a bifunctional enzyme will generate the correct peptides!

Title:Arg-C Title:dual ArgC_Cathep
Cleavage:R Cleavage:DX R
Restrict:P Restrict:P
Cterm Cterm

Arg-C definition Arg-C (N-term),
Cathepsin (C-term)
definition


DBTOOLKIT AND
DATABASE ON DEMAND


Working with databases: DBToolkit
http:/ / genesis.UGent.be/ dbtoolk it

See: M artens et al., Bioinform atics 2005, 21(17): 3584-3585


Summary of DBToolkit functionalities

a) Enzymatic digestion using regular or ‘dual’ enzymes
 proteins to peptides
b) N-terminal or C-terminal ragging
 enhancing the information content of the database
c) Non-lossy redundancy clearing
 raising database information ratio
d) Create shuffled and reversed databases
 false-positives testing
e) Extract sequence-based subsets
 a priori prediction of potential success rate
f) Map peptides back to proteins (maximal annotation approach)
 find all matching proteins, and select primaries
etc …


Database on Demand – DBToolkit online
http:/ / w w w .ebi.ac.uk/ pride/ dod

See: R eisinger et al., P roteom ics 2009, 9(18): 4421-4424


WHY DOES PROCESSING MATTER?


Serum degradation over time

From : Yi et al., Journal of P roteom e R esearch 2007, 6(5): 1768-1781


Plasma degradation over time

From : Yi et al., Journal of P roteom e R esearch 2007, 6(5): 1768-1781


TIME-LABILITY OF
SEQUENCE DATABASES


Example 1: HUPO PPP actualisation
Bringing the P P P from I P I 2.21 to I P I 3.13
1555 Total
1048 Unchanged 67%
507 Changed 33%
Of which:
338 Propagated 22% 67% (of ‘Changed’)
169 Defunct 11% 33% (of ‘Changed’)
Of which
95 Defunct (RFSQ_XP) 6% 56% (of ‘Defunct’)
Both exist, 72 Defunct (Ensembl) 5% 43% (of ‘Defunct’)
1 taxonomy now: RAT
1 immunoglobin
2 UniProt 0% 1% (of ‘Defunct’)

1048 + 345 = 1386 recoverable (89.1%)

See: M artens and M ueller et al., P roteom ics 2006, 6(18):5059-75


Example 2: human blood platelets
Bringing the P latelets from I P I 2.31 to I P I 3.13
673 Total
578 Unchanged 86%
95 Changed 14%
Of which:
78 Propagated 12% 82% (of ‘Changed’)
17 Defunct 3% 18% (of ‘Changed’)
Of which
5 Defunct (RFSQ_XP) 1% 29% (of ‘Defunct’)
12 Defunct (Ensembl) 2% 71% (of ‘Defunct’)

578 + 78 = 656 recoverable (97%)
See: M artens and M ueller et al., P roteom ics 2006, 6(18):5059-75


Proteins sometimes age badly

Adapted from : http:/ / w w w .ebi.ac.uk/ ipi


THE PICR MAPPING SERVICE


Identifiers through (name)space and time
http:/ / w w w .ebi.ac.uk/ tools/ picr

Limit search by
taxonomy
(pessimistic)
Submit accessions
OR sequences
(FASTA) with 500
entry interactive
limit (no batch
limit)
Choose to
return all
mappings or
only active ones

Select output format
Select one or
many databases
to map to in one
Run request
search

See: Côté et al., BM C Bioinform atics 2007, 8: 401


Mapping results


ESTIMATING FALSE DISCOVERY RATES
THE DECOY DATABASE APPROACH


Decoy databases, the latest fashion
Three main types of decoy DB’s are used:

- Reversed databases (easy)
LENNARTMARTENS  SNETRAMTRANNEL

- Shuffled databases (slightly more difficult)
LENNARTMARTENS  NMERLANATERTTN (for instance)

- Randomized databases (as difficult as you want it to be)
LENNARTMARTENS  GFVLAEPHSEAITK (for instance)

The concept is that each peptide identified from the decoy database is an incorrect
identification. By counting the number of decoy hits, we can estimate the number of
false positives in the original database.


Estimating the FDR (i)

2 × nbr _ decoy _ hits
FDR =
nbr _ forward _ hits + nbr _ decoy _ hits

FDR is the False Discovery Rate – it is a metric that gives you an indication of how
many (percent) of your identifications are potentially incorrect. Note that we multiply
the number of decoy hits by 2, because we should not only count the actual decoy
hits, but also the ‘hidden’ false positives that are present in the forward
identifications. The assumption here is that we expect one forward false positive hit
per decoy false positive hit, hence the doubling term.

From: Elias and Gygi, Nature Methods 2007, 4(3): 207-214


Estimating the FDR (ii)

nbr _ decoy _ hits
FDR =
nbr _ forward _ hits

This metric was proposed by Storey and Tibbs for genomics data, and further
investigated by Lukas Käll for proteomics. It provides a more accurate (and simpler!)
estimate of the FDR, but can be extended to also take into account the (suspected)
false positives in the forward set.

See: Storey and Tibbs, PNAS 2003, 100(16): 9440-9445
See: Käll et al,., JPR 2008, 7(1): 29-34


Thank you!
Questions?

�ݺ�ߣ

BITS - Overview of sequence databases for mass spectrometry data analysis

More Related Content

BITS - Overview of sequence databases for mass spectrometry data analysis