Position Masitinib specific scoring matrices were constructed to capture the interaction characteristics of SH3 domain families. A 3D QSAR method was used to predict the affinity of peptides on MHC. Physical energy terms, such as van der Waals, electrostatic, and desolvation energies, were used to predict the energy of domain peptide interactions. With these physical energy terms as feature vectors, machine learning techniques such as using a support vector machine were applied to peptide-SH3 domain interaction prediction. Statistical energy derived from the binding energy data and the complex structures was used to describe the specificity of SH2 domain. In other studies, protein-protein interaction data such as yeast two-hybrid was used to generate probabilistic models that can predict peptide sequences binding to SH3 domains. The development of prediction methods is needed because even current high-throughput techniques, such as SPOT, are limited to rather small sequence variations. Computational predictions for peptide-PRD can be applied to far more diverse peptides. Experimentally confirmed pre-publication binding data were kindly provided by Sachdev Sidhu at Terrence Donnelly Center for Cellular and Biomolecular Research, University of Toronto and Ben Turk at Department of Pharmacology, Yale University. The goal of the project was to predict the position weighted matrix, which defines the binding specificity of the target domains. Only the sequences of domains were given. We participated in the SH3 domain peptide specificity prediction category, for which five domains were given. In this way, each amino acid of the peptide can be assumed to have little interaction with the inter-peptide amino acids so that each position can be considered to be independent of each other, which is a necessary property for representing the binding specificity with the PWM. We approached this challenge by modeling the domain-peptide complex structures and then calculating the energy matrices describing the binding energy contribution of a specific amino acid at a specific position, excluding any prior experimental binding energy information. Fernandez-Ballester et al. tried a similar approach. In their research on the SH3 domain, they particularly focused on the generation of good template structures. However, peptides can have many conformations due to their thermal fluctuation. Thus, explicit consideration of structural ensemble would improve the binding energy prediction. In other previous researches, positions of peptides binding on PRD domains were considered as being fixed at the canonical motif. However, for general sequences without canonical motifs there is no clue how such methods can be applied. Furthermore, there is no guarantee that peptides in canonical motifs have the lowest energy. Thus it is worthwhile to test the effect of different sequence-structure mapping; in other words, the lowest energy binding position of peptide-domain needs to be investigated.