On the study of biomolecular interactions at different resolutions: Does size matter?

Biomolecular interactions are critical in cellular environments. Proteins, which are the workhorses of the cellular machinery, mediate by their interactions a wide range of cellular processes. Structural Biology is the scientific discipline concerned with revealing the molecular functions of these macromolecules through analysis of their three-dimensional structures. Classical structural biology techniques include X-ray crystallography, Nuclear Magnetic Resonance (NMR) and cryo-Electron Microscopy (cryo-EM). These experimental techniques have limitations that preclude their application to all biological systems. For example, large proteins (>50 kDa) are difficult to study by NMR spectroscopy and X-ray crystallography requires high quality crystals, which is not always trivial to achieve. For some systems, such as membrane proteins, their characterization by purely experimental techniques in their native environment is still challenging. Computational Structural Biology is a consolidated branch of science, whose goal is to understand the role that structure and dynamics play in the definition of the function of biomolecular systems. In particular, biomolecular interactions have been a major focus of this field over the past decades. For this purpose, various computational approaches have been designed and applied to the modelling of interactions, among which molecular dynamics- Monte Carlo-, docking- and, more recently, template modeling-based methods are the most widely used ones. Roughly, docking methods aim to build three-dimensional models of macromolecular structures by first, generating thousands of possible conformations (models), and then discriminating between biologically- and non-biologically-relevant models. Docking can be performed in the absence of any experimental information (ab initio) or by integrating information into the calculations (data-driven). In this thesis, several developments into the modeling of protein-protein and protein-nucleic acids interactions by computational integrative modeling approaches are presented. The thesis starts with a review of various representative models for downscaling the resolution of proteins, peptides and nucleic acids for the integrative modeling of their interactions (Chapter 1). These simplifications have two clear advantages: (1) It is easier to identify putative binding regions and (2) the computations become much more efficient. Real applications of the use of simplified model for modelling proteins and nucleic acids complexes are described, demonstrating that coarse-graining leads to more native-like models with a remarkable speed increase (Chapters 3 and 4). The implementation of information into the LightDock algorithm to both drive docking and scoring is described in Chapter 2. In this case, the use of experimental data such as mutagenesis data, translates into an increase in performance even when the data are not accurate and/or partially incorrect. In the final chapter (Chapter 5), several of the developments described in previous chapters (2 and 3) are combined for the modeling membrane-associated assemblies which are notoriously difficult to tackle. These systems are of special importance since are directly related to many diseases and therefore, are potentials target for drug design purposes. The thesis ends with a Conclusions and Perspectives section, giving a brief overview of chronological advances in both computing and Computational Structural Biology fields, together with some of the still open questions and challenges to be resolved in the near future.