Back to search

Institutes outside Greece

Integrative modelling of biomolecular complexes: from small to large

Abstract

dc:description

All cells, whether prokaryotic or eukaryotic, are finely tuned biochemical machines. In broad terms, genetic information is encoded in the nucleic acid sequence and is translated in functionally active biomolecules (proteins or other nucleic acids). These biomolecules then perform the multitude of functions the cell needs in order to maintain its homeostatic status. Biomolecules do not exist or perform their functions in isolation: They always act on – or together with – other molecules whether that is an enzyme catalysing a reaction involving a substrate, an activator protein acting on its target or a large collection of biomolecules coming together to create a large macromolecular machine such as the ribosome. Understanding cellular mechanisms in depth, therefore, requires understanding the makeup and function of these biomolecular complexes. For most types of complexes, truly understanding their function relies upon being able to obtain high-quality structures or models of the complex. Traditional structure determination techniques such as X-ray crystallography, Nuclear Magnetic Resonance (NMR) spectroscopy and cryo-Electron Microscopy (cryo-EM) have been used to determine the structure of thousands of biomolecules and biomolecular complexes. As of October 2019, the Protein Data Bank (PDB), the public repository of solved structures, counts more than 156000 entries. However, if one were to determine the unique entries in the database and then further focus on protein-protein complexes rather than free structures, the resulting number would only be a fraction of that (around 6000-7000 non-redundant biologically relevant complexes). With the protein-protein interactions in the cell estimated to be in the hundreds of thousands it quickly becomes clear that there is a significant gap between the number of biomolecular complexes with solved structures and the total number of complexes identified from interactome high-throughput studies. Next to the experimental methods mentioned above, another way of obtaining structural models for these complexes, the necessary step to understand the molecular mechanisms at play, is computational modelling. The field of computational modelling that deals with biomolecular complexes, which is the subject of this thesis, is integrative modelling, and in particular, biomolecular docking. These, like all subfields of computational simulation, share some of the same challenges, specifically sampling – or how to generate poses which resemble those of native complexes – and scoring – or how to identify good (or near-native) from wrong models in a large pool of models. Another challenge is about the way in which data are integrated into the simulations, or the need to weight those data in a way that allows for multiple data sources to be efficiently used in the same simulation while also reflecting the experimental uncertainties. This thesis focuses on two additional areas of interest: Docking of transmembrane TM protein complexes and protein-small molecule docking. Both areas are of great interest, both for academic and pharmaceutical research, as TM receptors constitute most drug targets and the majority of drugs on the market today are small compounds. The first half of this thesis pertains to membrane protein modelling, whereas the second half focuses on the modelling of protein-small molecule complexes. In the introductory Chapter, I provide an overview of the state of the integrative modelling field, of the ways in which data from diverse experimental sources can be used by modelling frameworks, and of recent advances in specific areas of interest. The thesis begins with the General Introduction which gently introduces some of the core concepts that are later expanded upon in the following chapters. It also briefly introduces the subject of each chapter. The main part of the thesis consists of Chapters 1 through 6. In Chapter 1, I provide an overview of the state of the field of integrative modelling with a particular emphasis on the types of data that can be used by integrative modelling frameworks such as HADDOCK (High Ambiguity Driven DOCKing), ROSETTA or IMP (Integrative Modelling Platform). The experimental methods that are discussed fall into one of three broad categories depending on the type of data that can be obtained from them: Interface-mapping techniques, techniques providing some kind of distance information between residues and shape-based techniques. Mutagenesis, HDX and NMR (when deriving chemical shift perturbations from titration experiments) are the interface-mapping techniques that are discussed, crosslinking, FRET and DEER the distance-based ones and cryo-EM and SAXS the shape-based techniques. For all these, I evaluate their relevance for the field of integrative modelling and provide examples of their application in modelling interesting and challenging targets. An additional focal point of this chapter is the evaluation of some computational methods in which significant progress has been made recently, namely the use of evolutionary information in the form of coevolution data in docking, advances related to the modelling of membrane proteins and applications of coarse-grained forcefields. In Chapter 2, I describe a recently published benchmark of membrane protein complexes. It is the first, and to the best of my knowledge, the only one of its kind, thus addressing a key missing element for further development of membrane protein docking algorithms. This non-redundant dataset consists entirely of transmembrane α-helical and β-barrel complexes, covering varying difficulty ranges from bound complexes (cases in which both bound components were extracted from the reference complex) to difficult, unbound cases with significant conformational rearrangements at the interface. Using this dataset, we define the baseline performance of HADDOCK for this type of complexes. In addition to the dataset itself we also make available a decoy set consisting of HADDOCK models produced during the benchmarking process. Chapter 3 is the last chapter which focuses on membrane protein modelling. In that Chapter I describe a protocol for HADDOCK, still under development, that implicitly represents the membrane bilayer by a shape consisting of layers of beads. Restraints are defined between these beads and Cα carbons of the subunits of the complex to drive them to the “membrane”. I compare the performance of these shape-restrained runs with one where a single centre-of-mass restraint is defined between the transmembrane segments of the two subunits. The performance of the shape runs is lower than expected compared to the simple transmembrane centre-of-mass restraint ones. Further work will be required to optimise this new approach. Chapters 4 through 6 constitute the second half of the thesis and revolve around protein-small molecule docking. In Chapter 4, I discuss our participation in a blind docking experiment – the 2016 iteration of the Grand Challenge experiment organised by the D3R consortium in which we had to model 36 protein-ligand complexes for the pharmaceutically relevant Farnesoid X receptor. Our small molecule docking protocol consists of the following steps: i) Identification of relevant protein receptor templates in the PDB and creation of an ensemble after clustering and selecting representative structures, ii) creation of a ligand ensemble after conformer generation and clustering of the resulting conformers and iii) docking using residues identified from the receptor templates. Despite excellent results for some cases, our overall performance was not so good compared to the other participants. We could identify the main limiting factor affecting the performance – namely the selection of the receptor template related to conformational changes taking place upon binding. Replacing the ensemble part of the protocol with the selection of a single template based on the similarity of its bound ligand with the target compound to dock does indeed leads to better results.In the following Chapter – Chapter 5 – I describe our participation in the following years’ Grand Challenge. In addition to selecting a single receptor template based on compound similarity we also revisit the conformer selection procedure: Ligand conformations for docking are selected based on their 3D shape similarity with the bound ligand present in the selected template. We also use the shape similarity to superimpose the selected conformers in the binding pocket of the receptor, bypass the initial stage of the docking protocol and directly proceed to refine the models using the water refinement stage of HADDOCK. The incorporation of shape information in our protocol has a significant impact on our success rate: Our submission (24 protein-ligand complexes predicted) was evaluated as one of the best within standard deviation of the top performing participant. Chapter 6 represents the logical conclusion of the small molecule docking part of the thesis. In that Chapter I describe a new HADDOCK protocol that makes use of the shape information which was identified as highly relevant in the preceding chapter. The innovative aspect of this protocol is the way the ligand shape is represented with the heavy atoms of the template compound being transformed into shape beads as used for the representation of the membrane described in Chapter 3. Similar to that protocol, restraints are then defined between the shape beads and the atoms of the generated ligand conformers. With the use of those shape restraints we don’t need to pre-select conformations, but instead use all 500 generated ones, increase the sampling and let HADDOCK select the near-native ones. This protocol outperforms the one described in the preceding chapter. The shape restraints allow to induce rather large conformational changes in the ligand toward its bound form. In addition, since we are now performing a full docking run, we can integrate any additional information in the simulation – something which was not possible in the previous protocol as it was a simple refinement. In the last Chapter – Chapter 7 – I summarise the main findings of this thesis and offer critical perspectives for the challenges the field is facing as well as some potential avenues worth exploring in the future.

Degree

thesis:*
Grantor dc:publisher
Institutes outside Greece
Year dc:date
2020

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Koukos, Panagiotis
  • Κούκος, Παναγιώτης

Subjects

dc:subject × 14

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/61122

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Koukos, Panagiotis; Κούκος, Παναγιώτης. Integrative modelling of biomolecular complexes: from small to large. Institutes outside Greece, 2020. http://hdl.handle.net/10442/hedi/61122