This server helps you in predicting the binding affinity of protein-protein complex of your interest.
We have developed the first sequence based method for predicting the binding affinity of protein-protein complexes using
a robust methodology based on the functional classification. We obtained a correlation and MAE of 0.91 and 0.52, respectively using jack-knife test.
Further, we have systematically analyzed the importance of selected features in each class and related with experimental observations.
It is evident that the percentage of binding site residues plays an important role in governing protein-protein binding affinity.
We suggest that our method (PPA-Pred) could be used as an efficient tool for protein-protein interaction network analysis for specific diseases
Further, we have refined our model
using the following criteria:
(i) selected the model using Akaike information criterion and early stopping set with limited number of three to five features,
(ii) developed the model with; inner feature selection, examined with outer leave-one-out cross-validation and tested with a blind dataset,
(iii) evaluated the performance using mean absolute error (MAE) and Pearson's correlation coefficient PCC), and (iv) assessed the statistical significance with p-values.
The refined method
could predict the binding affinity of 382(training)
protein-protein complexes with a MAE of 1.24 kcal/mol and 1.31 kcal/mol using leave-one-out cross-validation and blind test, respectively.
Dataset for 453 protein-protein complexes.
Prediction results for the training set of 382 complexes
Prediction results for the test set of 71 complexes
Click here to perform prediction