Swan River foreshore, Perth, Western Australia

University of Western Australia Biomedical, Biomolecular and Chemical Sciences

Ian Small Murray Badger David Day Harvey Millar

Steve Smith Barry Pogson Jim Whelan

AR

C C

en

tre P

lan

t En

erg

y B

iolo

gy

SUBASUBcellular location database

for Arabidopsis proteins

Sandra Tanz and Ian Castleden4th March 2011

Why protein localisation?

• Contributes towards the understanding of protein function and of biological inter-relationships, i.e. only proteins in the same location can interact.

• Separate subcellular locations often represent distinct cellular environments: proteins share similar attributes and play roles in defining the function of a subcellular compartment.

• To build hypotheses or models: large-scale phenotyping screens, microarray experiments and protein-protein interaction assays rely on protein localisation info.

How to localise proteins?

PredictionIn vitro uptake

(imports)

In vivo (GFP)Enzyme activity measurements

Western blot

Immunogold labeling

Subcellular proteomics (MS)

Protein-protein interaction

Images modified from Millar et al., 2009

SUBA: SUBcellular location database for Arabidopsis proteins

SUBA: SUBcellular location database for Arabidopsis proteins

What does SUBA document?

1193 5456942

MS (6398)GFP (2135)

SUBA II (2007) SUBA III (2011)

Combined sub-location data 250’719 1’022’040

Calls by PPI 0 6673

Calls by experiments (GFP, MS) 8273 19’528

Distinct proteins localised by GFP and/or MS 4531 8533

Bioinformatic predictions by 10 predictors 24 predictors

NEW!

Data mining

• Search of the NCBI PubMed (Medline) and Entrez (GenBank) databases using keywords

• Alert via Email

Data mining

• Search publication to extract localisation information = fully curated data

SUBA III interface http://suba.plantenergy.uwa.edu.au/

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III interface

SUBA III flatfile

Analysis of SUBA III data – on the way…

Do data become more or less consistent over time?

Experimental data (MS vs GFP)• How reliable are experimental localisation data? Has the overlap of

data changed with increasing data sets?

How reliable are GFP localisation data?

Total GFP localisations confirmed by MS

Total GFP localisations disputed by MS

1844 8306710

MS (9016)GFP (2554)

1386 73714458

MS (74172)GFP (1844)

1386 neither confirmed or disputed

Analysis of SUBA III data – on the way…

Do data become more or less consistent over time?

Experimental data (MS vs GFP)• How reliable are experimental localisation data? Has the overlap of

data changed with increasing data sets?• Does evidence for multiple locations mean the protein is dual

targeted/dynamic or is it a false positive?

Prediction vs experimental data• How reliable are predictors today?

PPI data• What do PPI data tell us about sub-cellular location? • Organellar proteome: Can we discover novel organellar proteins?

SUBA under the hoodSUBA under the hood

htt

p:/

/lib

rary

.du

ke.e

du

/dig

italc

olle

ctio

ns/

ge

dn

ey.

KY

01

80

/pg

.1/

• Why a Web interface?• GeneInvestigator, Mapman• AHM chemicals (Apache JPA)• For the foreseeable future databases are going to be

“Web” based (HTTP, Javascript, HTML ,css)• Need to be maintained by a minimum number of

developers (i.e. one!)

http://www.guistuff.com/

SUBA Tables (predictors)

SUBA Tables (“original” sources) http://www.ce4csb.org/amigo/

Suba Tables (publications)http://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?http://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&retmode=xml&id=18453549db=pubmed&retmode=xml&id=18453549

SUBA Tables (automation)

Julian Tonti-Filippini

Why Bother?

SELECT suba3.suba3.*, suba3.src_ppi_1.* FROM suba3.suba3 LEFT OUTER JOIN suba3.src_ppi AS src_ppi_1 ON suba3.suba3.locus = src_ppi_1.`locusA` WHERE EXISTS (SELECT 1 FROM suba3.src_ppi WHERE suba3.suba3.locus = suba3.src_ppi.`locusA` AND suba3.src_ppi.`locusB` IN (‘AT3G62420.1’))

“denormalisation” src_msms

Suba2

Suzanne M. Embury and Peter M.D. Gray

Computational Computational Systems BiologySystems BiologyCentre of ExcellenceCentre of Excellence

@suba.jsondef query(filter,offset=0,limit=1000): return Session().query(Suba3).filter(json2sqla(filter))\

.offset(offset).limit(limit)

http://suba.plantenergy.uwa.edu.au/cgi/suba.py/query?filter=['Suba3.ppi.locusB','in',['AT1G04234.1'],'AND','mwt','gt',80000.0]&offset=0&limit=1000

{success: True, result:[

{ locus:’AT1G54321.1’, mwt:81454, ….

ppi:[{locusA:’AT1G54321.1’,locusB:’AT1G04234.1’,pubmed:14567845}]},{ locus:’AT1G63021.1’, mwt:91454, ….

ppi:[{locusA:’ AT1G63021.1’,locusB:’AT1G04234.1’ ,pubmed:34567767}]},… ] }

Computational Computational Systems BiologySystems BiologyCentre of ExcellenceCentre of Excellence

(Near) Future

• Large number of predictors often given conflicting predictions… what to do?• Bayesian analysis…

Acknowledgements

Ian Small Harvey Millar

Joshua Heazlewood Julian Tonti-Fillipini

Thanks for your attention!!