AbstractsBiology & Animal Science

Protein functions are carried out through their interactions with other biological molecules. Research in protein interactions has attracted special interest from the scientific community for decades and still remains a hot research topic. Among others, there is an increasing interest to develop computational methods that automatically predict the 3D structure of protein-protein complexes, such as protein-protein docking. In this thesis, we propose novel approaches to assist protein-protein docking and protein similarity search. The proposed methods are based on the fact that when two proteins interact, their surfaces at the binding site demonstrate geometric complementarity (apart from physicochemical complementarity). The first docking method is based on geometric complementarity matching of the molecular surfaces. The basic idea of this algorithm is to use a descriptor that measures 3D shape similarity for computing shape complementarity, since equally-sized complementary surface patches tend to have similar shapes. A basic property of the proposed shape descriptor is its invariance to rotations of the surface patches. Complementarity matching is achieved through pairwise comparison of the local shape descriptors, thus, providing a fast geometric filtering and avoiding the exhaustive translational and rotational search of existing docking techniques.The second docking method extends this work in order to produce a docking algorithm that is robust to relatively small conformational changes of the interacting proteins. Since accurate surface complementarity has been proven to be inappropriate in unbound docking, the new method allows binding of surfaces with approximate surface complementarity. Experiments proved that the improved algorithm is more robust to conformational changes. Furthermore, a new scoring function is presented, which combines geometric complementarity with physicochemical factors, such as Coulomb potentials, van der Waals forces, hydrophobicity. The scoring function takes as input the list of candidate poses and a new rank list is produced having more near-native poses in the first positions. The last method that is proposed in this dissertation is an approach for molecular shape comparison. It aims to assist the problem of virtual screening, a process that is commonly used in rational drug design. More specifically, a new shape descriptor is proposed that combines local, global and hybrid local-global shape features. It is experimentally proven that the proposed compound descriptor is appropriate for similarity search of flexible ligands, while at the same time is robust in shape comparison of rigid proteins. Due to its compactness, the new descriptor enables fast screening of similar ligands to a target molecule from very large compound databases. Οι λειτουργίες των πρωτεϊνών πραγματοποιούνται κυρίως μέσω των αλληλεπιδράσεών τους με άλλα μόρια. Η έρευνα στο πεδίο των αλλλεπιδράσεων των πρωτεϊνών έχει προσελκύσει το ενδιαφέρον των επιστημόνων εδώ και δεκαετίες και εξακολουθεί να παραμένει ένα…