Studying protein-ligand interactions using a Monte Carlo procedure

Abstract

Biomolecular simulations have been widely used in the study of protein-ligand interactions; comprehending the mechanisms involved in the prediction of binding affinities would have a significant repercussion in the pharmaceutical industry. Notwithstanding the intrinsic difficulty of sampling the phase space, hardware and methodological developments make computer simulations a promising candidate in the resolution of biophysically relevant problems. In this context, the objective of the thesis is the development of a protocol that permits studying protein-ligand interactions, in view to be applied in drug discovery pipelines. The author contributed to the rewriting PELE, our Monte Carlo sampling procedure, using good practices of software development. These involved testing, improving the readability, modularity, encapsulation, maintenance and version control, just to name a few. Importantly, the recoding resulted in a competitive cutting-edge software that is able to integrate new algorithms and platforms, such as new force fields or a graphical user interface, while being reliable and efficient. The rest of the thesis is built upon this development. At this point, we established a protocol of unbiased all-atom simulations using PELE, often combined with Markov (state) Models (MSM) to characterize the energy landscape exploration. In the thesis, we have shown that PELE is a suitable tool to map complex mechanisms in an accurate and efficient manner. For example, we successfully conducted studies of ligand migration in prolyl oligopeptidases and nuclear hormone receptors (NHRs). Using PELE, we could map the ligand migration and binding pathway in such complex systems in less than 48 hours. On the other hand, with this technique we often run batches of 100s of simulations to reduce the wall-clock time. MSM is a useful technique to join these independent simulations in a unique statistical model, as individual trajectories only need to characterize the energy landscape locally, and the global characterization can be extracted from the model. We successfully applied the combination of these two methodologies to quantify binding mechanisms and estimate the binding free energy in systems involving NHRs and tyorsinases. However, this technique represents a significant computational effort. To reduce the computational load, we developed a new methodology to overcome the sampling limitations caused by the ruggedness of the energy landscape. In particular, we used a procedure of iterative simulations with adaptive spawning points based on reinforcement learning ideas. This permits sampling binding mechanisms at a fraction of the cost, and represents a speedup of an order of magnitude in complex systems. Importantly, we show in a proof-of-concept that it can be used to estimate absolute binding free energies. Overall, we hope that the methodologies presented herein help streamline the drug design process.

Las simulaciones biomoleculares se han usado ampliamente en el estudio de interacciones proteína-ligando. Comprender los mecanismos involucrados en la predicción de afinidades de unión tiene una gran repercusión en la industria farmacéutica. A pesar de las dificultades intrínsecas en el muestreo del espacio de fases, mejoras de hardware y metodológicas hacen de las simulaciones por ordenador un candidato prometedor en la resolución de problemas biofísicos con alta relevancia. En este contexto, el objetivo de la tesis es el desarrollo de un protocolo que introduce un estudio más eficiente de las interacciones proteína-ligando, con vistas a diseminar PELE, un procedimiento de muestreo de Monte Carlo, en el diseño de fármacos. Nuestro principal foco ha sido sobrepasar las limitaciones de muestreo causadas por la rugosidad del paisaje de energías, aplicando nuestro protocolo para hacer analsis detallados a nivel atomístico en receptores nucleares de hormonas, receptores acoplados a proteínas G, tirosinasas y prolil oligopeptidasas, en colaboración con una compañía farmacéutica y de varios laboratorios experimentales. Con todo ello, esperamos que las metodologías presentadas en esta tesis ayuden a mejorar el diseño de fármacos.