Modelling cofactors in comparative protein structure models by evolutionary inference

Proteins perform their role through the interactions they establish with other proteins and with small molecules, like ions or organic cofactors. The identification of these partners and of the mechanisms involved in their functional interactions can provide helpful insights into the molecular details of the protein annotation and for the development of new drugs. As many proteins lack of experimental structures and of annotated ligands, computational methods are required in order to predict these details and to guide the direction of experimental investigation.
In this context, our main aim is to enhance protein functional annotation and to improve comparative models by inferring their potential binding cofactors. Moreover, we want to evaluate the current state-of-the-art methods for binding site prediction in order to understand their advantages and limitations for future developments. Additionally, we aimed to improve the assessment of binding site prediction methods by creating an automated system of continuous model evaluation. Finally, we created a new binding site descriptor for the de novo ligand and binding site prediction in protein models.
The content of this thesis is organized as follows. Chapter 1 introduces protein structure, binding sites and experimental techniques for structure determination; moreover, we illustrate the current approaches to model protein structures and to predict their ligand binding sites. In chapter 2, we describe the assessment of the ligand binding site predictions within the 9th edition of the Critical Assessment of protein Structure Prediction (CASP) experiment, while in chapter 3 we discuss the latest developments in the 10th round. Within chapter 4 we illustrate the evolution of this assessment into the Continuous Automated Model EvaluatiOn (CAMEO) Ligand Binding category and we describe the homology predictor, which is used as reference for the comparison of the other methods registered to CAMEO. Chapter 5 presents the new SWISS-MODEL server, which employs a base ligand modelling pipeline to place potential small molecules partners, inferred from the target’s template, into the built models. Motivated by the performances of the previous method and by the results seen in the last CASP editions, in chapter 6 we present a new method to model ligands, especially ions and organic cofactors, into comparative models; this approach is based on the analysis of the similarities between a target and its homologous proteins. In chapter 7, we describe a novel descriptor for ligand binding sites, based on moment invariants and developed for the de novo prediction of ligands. Finally, in chapter 8 we draw the general conclusions of the work presented in this thesis.