Proteome scanning of PDZ domain interactions using support vector machines
Shirley Hui and Gary D. Bader
BMC Bioinformatics 2010, 11:507 (12 October 2010)
An accurate predictor of genomic PDZ domain interactions would allow the proteomes of organisms to be scanned for potential binders. Such an application would require an accurate and precise predictor to avoid generating too many false positive hits given the large amount of possible interactors in a given proteome. Once validated these predictions will help to increase the coverage of current PDZ domain interaction networks and further our understanding of the roles that PDZ domains play in a variety of biological processes.
We built an SVM using mouse and human experimental training data to predict PDZ domain interactions. We showed that it correctly predicts known interactions from proteomes of different organisms and compared to published state of art predictors, is more accurate and precise.
SVM predictions were validated using known interactions from PDZBase, a domain peptide interaction database. To further support our predictions, the number of interactions which also corresponded to known protein-protein interactions (PPIs) was calculated for 213 PDZ domains with interactions from iRefIndex. iRefIndex is a PPI database which consolidates PPIs from different databases including BIND, BioGRID, CORUM, DIP, HPRD, IntAct, MINT.
The following are SVM proteome scanning predictions for 13 human, 6 fly and 6 worm PDZ domains with domain-peptide interactions in PDZBase.
The following are SVM proteome scanning predictions for 192 human PDZ domains for which the SVM predicted binders. From this set 75 PDZ domains had predicted interactions which corresponded to PPIs in iRefIndex. Please see Supplementary Information for more details.
The format of the output is:
<indicator> <predicted binder sequence> <decision value> <source> <transcript ids>
<indicator> is one of the following symbols:
- * = validated by PDZBase or corresponds to an iRefIndex PPI
- X = false positive as determined by protein microarray experiments (only for fly and worm)
- empty = no experiment or other evidence to validate or support this prediction
<predicted binder sequence> is the sequence of length five of the predicted binder
<decision value> is a real number computed by the SVM to evaluate if a given sequence should be predicted as a binder or not. All values will be greater than zero since the files only contain predicted binders.
<source> is non empty if only indicator is non empty and is one of the following codes:
- PB = found in PDZBase
- IR = corresponds to a PPI in iRefIndex (transcript index1, ..., transcript index n)
- Ensembl TRS ids with tails corresponding to the predicted binder
- Chen model parameter and binding site encoding files
- Stiffler model parameter files
- Ensembl proteome files for Human, Worm and Fly
- Experiment Interaction files (in peptide file format)
- Fly files from Chen
- Human files from Sidhu
- Mouse files from Stiffler
- Worm files from Chen
- Curated Interaction files (flat files)
- PDZBase for Human (Worm and Fly included, but not used)
- iRefIndex interactions for Human
- Phage codon bias files
Files required to run the ProteomeScan software
- jfreechart 1.0.12 (and dependencies)
- weka 3.9.1
auc calculator (Davis & Goadrich, 2006)
- iText 2.1.3
- Bingo 2.3
- Cytoscape 2.6.3
- Cytoscape-task 2.6.3
- BRAIN 1.0.5 (pdzsvm)
- libSVM 2.8.9 (pdzsvm)
- Shirley Hui
- Gary Bader