PepSite
A structural method to predict peptide/protein binding
Evangelia Petsalaki, Alexander Stark, Eduardo Garcia Urdiales, Rob Russell
A structural method to predict peptide/protein binding
Evangelia Petsalaki, Alexander Stark, Eduardo Garcia Urdiales, Rob Russell
Tips and Tricks
A. If you have a protein structure and don't know what it interacts with and how
B. If you have a protein structure and what it interacts with but not how and with what peptide
C. If you know that one protein of known structure interacts with another by a linear stretch that is long but you don't know what part of it or where it binds
Results
Table 1. Sensitivity on the 232 SCOP families of peptide-protein complexes of PepSite using different cutoffs for specificity when tested against itself (self consistency), with jack-knife cross validation , and the same on the 195 SCOP families of unbound structures of the dataset.
Table 2. Sensitivity on the 450 peptide-protein complexes of PepSite using different cutoffs for specificity when tested against itself (self consistency), with jack-knife cross validation , and the same on the 405 unbound structures of the dataset.
A. If you have a protein structure and don't know what it interacts with and how
- Use STRING to identify interacting partners
- Right click on your protein and run Dilimot to identify potential binding motifs
- Use the peptides predicted by Dilimot as input peptides for PepSite to see where they bind and the score for binding
B. If you have a protein structure and what it interacts with but not how and with what peptide
- Use STRING to plot the interacting partners or use Dilimot directly
- Right click on your protein and run Dilimot to identify potential binding motifs
- Use the peptides predicted by Dilimot as input peptides for PepSite to see where they bind and the score for binding
C. If you know that one protein of known structure interacts with another by a linear stretch that is long but you don't know what part of it or where it binds
- Split the linear stretch in windows of 10-residues-long peptides and run PepSite with these (If your linear stretch is too long for such a study contact us)
- Make a graph of the p-values and average them in windoes of 10 residues so that the graph is smoother and more informative
- The regions with lower p-value (and if it is lower than 0.1) have the biggest probability for interacting with your protein
Results
Table 1. Sensitivity on the 232 SCOP families of peptide-protein complexes of PepSite using different cutoffs for specificity when tested against itself (self consistency), with jack-knife cross validation , and the same on the 195 SCOP families of unbound structures of the dataset.
specificity | self consistency | jack-knife | unbound | unbound jack-knife |
0% | 66.00% | 63.11% | 60.49% | 59.26% |
60% | 41.78% | 41.78% | 34.32% | 30.06% |
70% | 31.33% | 32.44% | 29.14% | 29.63% |
80% | 23.78% | 26.67% | 20.74% | 23.20% |
90% | 13.11% | 17.56% | 10.62% | 14.81% |
98% | 3.11% | 6.22% | 1.73% | 5.93% |
Table 2. Sensitivity on the 450 peptide-protein complexes of PepSite using different cutoffs for specificity when tested against itself (self consistency), with jack-knife cross validation , and the same on the 405 unbound structures of the dataset.
specificity | self consistency | jack-knife | unbound | unbound jack-knife |
0% | 64.22% | 64.66% | 63.59% | 61.54% |
60% | 42.67% | 43.53% | 44.10% | 43.59% |
70% | 34.05% | 36.21% | 36.92% | 36.92% |
80% | 29.74% | 36.21% | 31.28% | 32.82% |
90% | 17.24% | 20.26% | 17.44% | 23.08% |
98% | 4.31% | 9.48% | 4.10% | 11.79% |