Joan Francesc Gilabert Navarro defends his thesis on computational methods for drug development

Great advances in computing power have motivated the hope that computer simulation methods can accelerate the pace of new drug discovery. For this to be possible, fast, accurate and easy-to-use tools are needed. One of the problems that has received the most attention is the prediction of protein-ligand binding free energies. Two major problems have been identified for these methods: lack of sampling and model approximations.

This thesis is focused on solving the first problem. To this end, we present the development of efficient methods for the estimation of free energy binding between protein and ligand. We have developed a protocol that combines methods called enhanced sampling with classical simulations for greater efficiency. Enhanced sampling methods are a class of tools that apply some type of external perturbation to the system being studied in order to speed up sampling.

In our protocol, we first run an exploratory simulation of enhanced sampling, starting with a sample of the binding of the protein and the ligand. This simulation is partially biased towards those states of the system where the two components are more separated. We then use the information obtained from this first shorter simulation to run a second longer simulation, with unbiased methods to obtain a reliable statistic of the system.

Thanks to the modularity and the degree of automation that the implementation of the protocol offers, we have been able to test three different methods for long simulations: PELE, molecular dynamics and AdaptivePELE. PELE and molecular dynamics have shown similar results, although PELE uses fewer resources. Both have shown good results in the study of fragment or protein systems with inflexible binding sites. However, both have failed to reproduce the experimental results for a kinase, Mitogen-activated protein kinase 1 (ERK2). On the other hand, AdaptivePELE has not shown much improvement over PELE, with positive results for the Urokinase-type plasminogen activator (URO) protein and a clear lack of sampling for the progesterone receptor (PR).

In this work we have demonstrated the importance of establishing a balanced test bed during the development of new methods. By using a diverse test bench we have been able to establish in which cases the protocol can be expected to obtain accurate results, and which areas need further development. The test bench has consisted of four proteins and more than thirty ligands, many more than those commonly used in the development of methods for the prediction of binding energies using pathway-based methods.

In summary, the methodology developed during this thesis may contribute to the process of researching new drugs for certain types of protein systems. For the rest, we have observed that unbiased simulation methods are not efficient enough and more sophisticated techniques are needed.