Desenvolvimento de funções empíricas quântico-clássicas para a predição da afinidade proteína-ligante

• Autor(a) principal: Borsato, Nicolau Gonçalves
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do LNCC
• Texto Completo: https://tede.lncc.br/handle/tede/359
• topic: Atracamento molecular; QM/MM; DockThor (Programa de computador); Equação de Schrödinger; CNPQ::CIENCIAS BIOLOGICAS::FARMACOLOGIA

Molecular Docking is a methodology widely used in Structured Based Drug Desing (SBDD). And it has two main objectives: i) ligand pose prediction (i.e. predict the ligand bind mode) and ii) bind affinity prediction of a small molecule in a selected receptor’s active site. It is an approach widely used in virtual screening, aiming the identification and the optimization of prototypes composts and leaders (possible future drug candidate). To perform these objectives, mainly the objective ii), it’s needed the use of accurated evaluation functions. Nowadays, evaluation functions have good performance in identify these binding modes, in contrast with an accuracte affinity function, ranking in virtual screening experiments and the classification of active and inactive ligands, constitutes huge challenge. Recently, it was developed in our group GMMSB/LNCC empiric functions (DockTScore) to predict the protein-ligand bind affinity. These functions are based in the force field MMFF94s and physical-chemical descriptor associated with the protein-ligand binding. These functions incorporated to the program and webportal DockThor-VS (www.dockthor.lncc.br). The functions developeds have shown highly competitive with others published functions, but yet they need improvement, principally in stronger protein-ligand binding affinity s (i.e., Ki < 45 nM). There is an hypothesis that complexes exhibiting strong bind affinity they are associated with specific interactions (e.g halogen bind, polarization, interaction with metal atoms pi and T-stacking, etc.) which will hardly be well described using a classical force field and they need to be evaluated by eletronic structure quantum calculations or hybrid such as QM/MM. In this work, it was developed a protocol for preparation and evaluation of protein-ligand complexesby QM/MM using NAMD and MOPAC, aimed to deveolp an empirical hibrid quantum-classical function. Also are objectives of this work: i) identify which are the best classical descriptors in the function DockTScore to be used with the protein-ligand interaction energy calcualted by an hybrid QM/MM method; and ii) compare the outcomes of a semi-empirical hamiltonian PM7 and PM6D3H4X. The linear empirical functions was developed using the set "PDBbind v2013 refined set" containing 2764 protein-ligand complexes. The "coreset" containing 195 complexes was used as an independent test and the results compared with the results obtained by DockTScore function. For the realization of this work it was necessary the development of an automated protocol to settle topologies to recognize residue protonations (i.e., ASP e GLU). It was investigated and compared protocols with and without energy minimization 15A QM. The results obtained shows that the interaction energy of quantum calculations correlates better with the affinity data (∆G) obtained through experiment than with classical force field based data. The results also indicates that the usage of the energy minimization process can be very useful to obatin better correlations, principaly when halogen interaction is involved and metal cofactors.