This web page contains supplementary information for the paper "Graphlet
kernels for prediction of functional residues in protein structures",
by Vladimir Vacic,
and Predrag Radivojac,
Journal of Computational Biology. 17(1):55-72. (2010)
Preliminary version of the manuscript can be downloaded
Supplementary Table S1: A nonredundant
subset of phosphorylated sites mined from PDB and results of the search
for the structures of proteins that can be found both in the phosphorylated
and the unphosphorylated states.
Supplementary Table S2: Summary of the set of sequences with experimentally annotated
phosphorylation sites. NR: non-redundant, here defined as having less than 40% sequence identity in the
25 residue-long fragment centered at S, T or Y.
Supplementary Table S3: Performance of method/parameter combinations on the CSA dataset.
Within each group of predictors, the one with the highest AUC was awarded a point. In the case of ties, the
point would be evenly split between the methods which performed equally well.
Supplementary Table S4: Performance of method/parameter combinations on the PHOS dataset.
Within each group of predictors, the one with the highest AUC was awarded a point.
Supplementary Table S5: Jaccard coefficients between sets of edges in residue interaction networks
obtained using different methods. Jaccard similarity coefficient for two sets A and B is defined
as J(A,B) = |A \ B|/|A [ B|. Values over 70% are in bold face and marked with an asterisk.
Supplementary Figure S1: Schematic representation
of the BLOSUM50 matrix-based amino acid alphabet reduction.
Supplementary Figures S2-S24: Performance
of all predictors on the CSA and PHOS datasets.
Beta version of the graphlet kernel code is available for