Analysis and prediction of protein subcellular localization for Gramnegative bacteria

1
Analysis and prediction of protein subcellular
localization for Gram-negative bacteria

• Jennifer Gardy
• Dept. of Molecular Biology Biochemistry
• Simon Fraser University
• Burnaby, B.C.

2
Gram-negative Subcellular Localization
3
PSORT-B www.psort.org/psortb

• Web-based subcellular localization prediction
  tool
• Analyzes 6 biological features using 6 modules
• More comprehensive than existing tools
• Score for each of 5 primary Gram -ve localization
  sites
• PSORT I does not predict extracellular proteins
• Designed for high precision (97) (specificity,
  )
• PSORT Is specificity measured at 59
• Trained and tested using a dataset of proteins of
  experimentally-verified subcellular localization
• Constructed manually through literature review
• Largest dataset of its kind
• Freely available at the PSORT-B site

4
PSORT-B 101
Signal peptides Non-cytoplasmic Amino acid
composition Cytoplasmic Transmembrane helices
Inner membrane PROSITE motifs All
localizations Outer membrane motifs Outer
membrane Homology to proteins of known
localization All localizations
5
Understanding The Results

• Output available in several formats
• NCBI Genomes have been pre-computed
• www.psort.org/psortb/genomes

6
Current Limitations

• Sensitivity is not emphasized in this version
• Will not always get a prediction
• Examine your results carefully!
• Proteins at multiple localization sites
• Flagged in comments, score distribution
• Certain classes difficult to identify
• Inner membrane with 1-2 helices
• Extracellular
• Lipoproteins not identified in this version
• Trained primarily on proteobacteria
• Reduced predictive ability for bacteria with
  atypical cell walls
• Gram-negative bacteria only
• Use PSORT I for Gram-positives

7
Insights Gained To Date

• Localization is an evolutionarily conserved trait
• SCL-BLAST (specificity of 96.7, sensitivity of
  60.4)
• E-value cutoff of 10e-10
• Length restriction HSP 80-120 length of query
  to avoid matches to single domains
• Identified motifs characteristic of outer
  membrane proteins through a data mining approach
• 279 sequences frequent in OMPs, infrequent in
  non-OMPs
• Typically 6 residues long, may occur in
  combinations
• Used in OMP classifier
• PSORT-B v.1.1 (3 motifs) spec. 100, sens. 24
• SVM approach (for next version) spec. 98
  sens. 81
• Motifs map primarily to periplasmic turn regions
  of known 3D structures
• May reflect importance of periplasmic turns in a
  transmembrane beta-barrel structure vs. other
  similar non-membrane barrel structures

8
Future Directions

• Development of a Gram-positive version
• Dataset being constructed (Dr. S. Rey, SFU)
• Increasing existing Gram-negative dataset
• Literature review, text mining
• Improvement of Gram-negative modules
• New computational techniques
• New biological information

9
Still Curious?

• www.psort.org/psortb - Documentation, Users
  Guide, datasets motifs used in the program,
  many other subcellular localization prediction
  resources
• PSORT-B is described in
• J.L. Gardy et al (2003). PSORT-B improving
  protein subcellular localization prediction for
  Gram-negative bacteria, Nucleic Acids Research
  31(13)3613-17
• The data mining approach to OMP motif discovery
  is described in
• She, R, Chen, F., Wang, K., Ester, M., Gardy,
  J.L, and F.S.L. Brinkman (2003). Frequent
  Subsequence-Based Prediction of Outer Membrane
  Proteins. 9th ACM SIGKDD International Conference
  on Knowledge Discovery and Data Mining.

10
Acknowledgements
Initial PSORT-B Development Fiona S.L. Brinkman
Cory Spencer Martin Ester Ke Wang Gábor E.
Tusnády István Simon Katalin deFays Christophe
Lambert Sujun Hua Kenta Nakai Ongoing PSORT-B
Work Sébastien Rey (SFU) Matt Laird
(SFU) Also Bob Hancock (UBC) Oliver Schulte
(SFU) Søren Brunak (CBS) Gunnar von Heijne
(Stockholm) Funding Natural Sciences
Engineering Research Council
www.pathogenomics.sfu.ca/brinkman