Integrative Modelling of Biomolecular Complexes: From Small to Large

Chapter 1 provided a detailed and comprehensive review on the types of data than can be used by Integrative Modelling software like HADDOCK, ROSETTA and IMP, with a particular emphasis on the experimental techniques which can be used to map interfaces, derive distance restraints or shape-based approaches. Another focal point of the chapter is how recent advancements have affected the field of membrane protein modelling. Chapters 2 and 3 also relate to membrane protein modelling with the former describing a recently available benchmark comprised of ready-to-dock membrane protein complexes and the baseline performance of HADDOCK for the entries of the benchmark, and the latter, ongoing work regarding development of a protocol for HADDOCK for the docking of transmembrane protein complexes. The remaining of the thesis focused on small molecule modelling with Chapters 4-6 detailing three separate protocols for the docking of small molecules and protein receptors, with every protocol and chapter reflecting methodological improvements over the previous one. In Chapter 4, I described the participation of HADDOCK in the 2016 iteration of the Grand Challenge, the blind docking experiment organised by the D3R consortium. While our performance in the pose prediction component was not impressive, we could identify the main factor limiting HADDOCK’s performance, namely the selection of appropriate templates for the receptor and came up with an improved way of selecting receptors. Chapter 5 described additional improvements in our protocol related to the way the compound conformers are selected prior to docking which led to our participation in the 2017 iteration of the Grand Challenge being evaluated as one of the best. Chapter 6 detailed the development of a new protocol for protein-small molecule docking, by combining the lessons and conclusions from Chapters 4-5 and formalising their approaches in a method that relies on HADDOCK’s main strength, its ability to incorporate information to guide the simulation. This new, shape-restrained docking protocol outperformed all our previous efforts while at the same time not relying on any external software. A common denominator between the membrane protein work and the small ligand docking discussed in this thesis is the use of shape information. Indeed, chapters 3 and 6 both describe applications of shape information represented as beads to drive the modelling process. In Chapter 3 one or more layers of beads are used to implicitly represent the membrane and in Chapter 6 ligand docking is restrained to a shape based on the structure of a homologous compound. Despite the commonalities between the two protocols, the outcome of the docking is very different between the two, with the small molecule protocol achieving high-quality results and improving upon our previous efforts in this area, whereas the membrane one achieves results which are only marginally better than defining centre-of-mass restraints between the transmembrane segments of the partners for the docking. A main limiting factor in the case of membrane protein complexes seems to be the size of the complex, which defines the number of restraints defined between shape and molecules and negatively impacts performance.