The document describes Representative Proteomes (RPs), which provide a standardized, stable, and unbiased set of proteomes. RPs are generated computationally based on pair-wise comparisons of proteomes and are inspected by curators. RPs reduce the number of proteomes by over 90% while retaining over 95% of protein families and experimentally characterized proteins. RPs integrate with UniProt/UniRef and are released monthly. They allow mapping of proteomes to reference genomes and aid in sequence analysis and metagenomic sequence classification.

1. Representative Proteomes
and Genomes
A standardized, stable and unbiased set of proteomes and
genomes
http://pir.georgetown.edu/rps/
Raja Mazumder (mazumder@gwu.edu)

2. Nomenclature
 Representative Proteomes
 Primarily computational
 Reference Proteomes/QFO
Proteomes/Blessed Proteomes
 Primarily manual
 Extension of Reference proteomes

3. Representative Proteomes (RP) and Representative
genomes (RG)
http://pir.georgetown.edu/rps/

4. Procedure to generate the RPs
Compute pair-wise co-membership value (X) in UniRef50 for all proteomes
For each proteome, compute the mean co-membership between
this proteome and the other proteomes
Create ranked proteome list based on the mean co-membership
RPG generation starts
For a given CMT, take the first proteome in the ranked list and
the ones with X ≥ CMT to form an RPG, and remove them from
ranked list
ranked
No
list
empty?
Yes
RPG generation ends
Select a Representative Proteome for each RPG (manually inspected by curator)

5. Proteome A Proteome B UniRef50
UniRef50
 Sequence clusters (UniRef100, 90, 50)
 From any organism
 Part of UniProt production cycle
 PMID: 17379688

6. RPs at Different CMTs
1000 100
900 90
3.02
800 80
700 2.69 70 # RPs
Million proteins
600 2.36 60
# RPs
% Reduction -
500 50 % 1
Proteomes
400 40
2.02 % Reduction -
300 30 Sequences
2
200 20
100 10
0 0
75 55 35 15
1
Based on 1144 complete genomes
CMT (%) 2
Based on 4.3 million sequences (complete genomes only)
UniProtKB total: 13.46 million sequences

7. • RP at higher level is used to cluster the lower levels
• RPGs are constructed based on co-membership, not
taxonomy

8. Manual mapping of UniProt and
NCBI genomes
 The taxonomy ID of each proteome present
in UniProt is mapped to the NCBI
RefSeq/GenBank genome project IDs
 When more than one genome is available
for the same taxonomy ID, the genomes are
ranked according to the availability of a
RefSeq genome, number of related
publications, number of citations for each
publication, and date of sequencing.
 The highest ranking genome is mapped to
the UniProtKB proteome

9. RefSeq genomes and proteomes
 Mapping allows us to retrieve genomes and
proteomes from RefSeq.
ftp://ftp.pir.georgetown.edu/databases/rps/rg
ftp://ftp.pir.georgetown.edu/databases/rps/rp_in_refseq_sequences/

10. RP55 Over Time
1400 120 # complete proteomes
# RPGs
% species in multiple RPGs
1200
100 % stable RPGs
1000
# Complete proteomes
80
800
60 %
600
40
400
20
200
0 0
2004_1 2005_4 2006_7 2007_10 2008_13 2009_15 2010_09
UniProtKB release

11. Coverage Statistics – RP55
 95% of all InterPro families contain at least
one protein from the RP set
 InterPro covers ~75% of all proteins in
UniProtKB, and this number holds true as
well for RP55
 93% of the experimentally-characterized
proteins are retained in the RP set

12. Coverage and use

13. Downloads

14. Make your own set

15. Visualizing all-against-all proteome
correlation matrix vs. the taxonomy
tree
 Developed a method to graphically visualize NCBI's taxonomy
tree and overlay the proteome correlation tree (PCT) to illustrate
genomic similarity between organisms that may otherwise be
considered to have distant ancestry.
 Computed all-against-all correlation values between all complete
proteomes
 Comparison network can be browsed in Cytoscape network
software to easily identify nodes in the taxonomy tree that are not
supported by PCT data
 Development tools: CytoscapeWeb, CytoscapeRPC, Perl

16. Family
Enterobacteriaceae Distance based on taxonomy tree
Shigella Escherichia Enterobacter Klebsiella Genus
Distance based on taxonomy tree
Species
ENT38 Distance based on correlation table
ECOLI
ESCF3 ECO24
ENTAK
ECOK1 KLEP7
SHIFL
SHIF8

17. Example: Examine correlation scores of
AGRT5
 Agrobacterium tumefaciens (AGRT5)
http://pir.georgetown.edu/cgi-bin/rps_tree.pl?point_id=r15p176299&on=1&on100=1&file_id=122063&p=#-5

18. Can easily identify genomic neighbors
 The top 2 levels are family
and genus nodes arranged
according to taxonomic
position
 The bottom nodes are
complete proteomes with a
heuristic force-directed
layout applied according to
all-against-all correlation
 Although AGRT5 and
AGRVS share the same
genus, they are relatively
distant from each other
(~28%), compared to
AGRT5 and AGRSH
(~70%).

19. Sequence search
 Cleaner BLAST/phmmer results

20. Conclusions
 High quality RPs generated computationally and
inspected by curators
 A standardized, stable and unbiased set of proteomes
and genomes
 Completely integrated and into the UniProt/UniRef
production pipeline and has monthly releases
 Automatically selects QFO/UniProt RF (if available in
RPG) as RP (provide feedback to QFO and others if
discrepancy)
 Extended to RefSeq (
ftp://ftp.pir.georgetown.edu/databases/rps/rg;
ftp://ftp.pir.georgetown.edu/databases/rps/rp_in_refseq_s
equences/)
 RGs can help placement of unknown metagenomic
sequences into the correct clusters

21. Acknowledgements
Chuming Chen (PIR)
Darren Natale (PIR)
Hongzhan Huang (PIR)
Jian Zhang (PIR)
Peter McGarvey (PIR)
Cathy Wu (PIR)
Mona Motwani (GWU)
Jamal Theodore (GWU)
Robert Finn (HHMI Janelia Farm/Pfam)
Eleanor Stanley (EBI)
Kim Pruitt (NCBI)
Yuri Wolf (NCBI)
UniProt Consortium