Prediction of protein-protein interaction sites using information from protein sequence and 3D structure is purposed in this thesis. The protein interaction site is defined as a slide window of 9 residues in which at least central residue is in contact with other chains. The aminoacid is in contact if its distance to the nearest heavy atom of a neighbour chain is less then 6 A. In this work it is used an existing non-redundant dataset which is classified using Random Forest and OR method. OR method, purposed in this thesis, is a method based on combining classifiers and improvement of F-measure in cases when data is imbalanced. The predictions are based on sequence alone or combination of sequence and 3D structure features. The following 3D features are used: sequence, non-polar ASA, maximal depth index, average depth index, minimal protrusion and hydrophobicity. The results obtained using prediction based only on sequence information are better for other in the field, while the results based on both sequence and 3D structure information is the rank of results of other authors. In addition OR method was tested on 9 UCI datasets, where it achieves improvement of the results comparing with methods based only on individual classifiers. Furthermore, using the information about the potential binding site determined by above mentioned methods the wild type Ras and the RBD (Ras Binding Domain) of C-RAF1 were docked (by specifying the predicted binding site as a centre). Finally the obtained complex was energy minimised and subjected to molecular dynamics.