Resumen de Parmbsc2: development of an accurate force field for nucleic acids simulations
Molecular dynamics (MD) simulations are the equivalent to a computational microscope allowing us to interrogate molecules and their behaviour in complex systems. They are recognised as a valuable and accurate tool in reproducing experimental evidence and predicting biomolecules flexibility at a resolution level which experimental techniques can’t usually reach. Running a MD simulation implies the propagation of Newton’s equations of motion of a system composed by beads connected by bonds. The core of MD is the energy functional: the force field, a classical Hamiltonian, defined by a set of empirical parameters defining how the energy of the system change as a consequence of geometrical alterations. Highly accurate force fields are nowadays available for proteins, nucleic acids, membranes, organic and inorganic ligands, materials, and many other systems. Focusing on nucleic acids, state-of-the-art force fields for DNA (parmbsc1 and OL15) are now able to predict B-DNA structure and flexibility with the same accuracy as NMR and X-ray experiments, and correctly reproduce unusual secondary motifs, alternative DNA forms, DNA binding with numerous partners, folding and melting processes, etc. On the contrary, RNA structure and flexibility is more difficult to capture, and the maturity of RNA force fields is far from being as “good” as the one reached for DNA, and consensus exists that a new force- field able to capture the rich conformational freedom of RNA should be developed. A substantial part of this thesis focuses on this precise problematic. This thesis is transversal to many fields, going from applied structural bioinformatics to pure physics-based modelling using classical and quantum approximations. We created the R package veriNA3d to systematize the process of handling PDB contents in diverse formats, including facilities to analyse large structures like ribosomes. VeriNA3d showed to be particularly useful for extracting detailed information on DNA-protein interactions, analysing any desired conformational space sampled experimentally, or generating, as for the purpose of this thesis, a specific dataset of all stacked nucleobase arrangements (over 3x105 structures). To reach a deep understanding of stacking interaction at both structural and energetic level, high-level QM approximations were used, and a Machine Learning algorithm was developed that showed to accurately predict expected stacking energies. We also proposed a new set of non-bonded parameters in Lifson’s force field framework, meant to be part of the future parmbsc2. Parmbsc2 is currently under beta- testing for both DNA and RNA systems. The thesis is written in monographic form. It covers a range of topics in nucleic acid structure and modelling introduced in chapter 1 with a general overview. The introduction leads to the objectives of this work in chapter 2. Chapter 3 focuses on methodological details covering quantum mechanics, molecular mechanics, and a detailed explanation of the force field parameterization scheme for non-bonded terms and dihedral angles. Chapter 4 is reserved for the results, and chapter 5 for the general discussion. The chapter 6 closes this thesis with the conclusions.
