There is a gap between our detailed knowledge of macromolecules present in many organisms, as provided by genome sequencing, and our scarce comprehension of the function of biological systems. A fundamental reason is that the functional units of the cell are large macromolecular complexes, such as the ribosomes, proteasomes, or metabolons, whose detailed structure and dynamics are in many cases unknown. A clear understanding of protein-protein interactions, both for stable complexes and for transient ones, would therefore mark a major step forward, and it would have enormous consequences for designing new strategies of therapeutic intervention in diseases like cancer. This understanding can be realistically achieved only by a successful combination of experimental and computational techniques. In facts, often limitations in the availability of high-quality samples for X-ray crystallography call for the use of lower-resolution techniques like cryo-EM, where a compelling reconstruction of the structure of the complex requires the combined use of advanced computational techniques. The same is true for the problem of interpreting the dynamics of macromolecular aggregates from NMR spectroscopy, especially in the presence of natively unstructured proteins.
When the structure of the interacting proteins is known, either from experiments or from modeling, several docking algorithms are available which attempt to predict the geometry of the complex. However there is considerable room for improvement, as monitored e.g. by the periodic Critical Assessment of Predicted Interactions (CAPRI) competitions. Furthermore, despite a number of simulations appearing in the literature, the detailed dynamics at atomistic level of the association/dissociation process and its thermodynamic and kinetic quantification remains still unclear: besides direct electrostatics and dispersion forces, what is the contribution of water-mediated interactions and conformational changes? Can we obtain actual binding pathways, e.g. does molecular dynamics entail relevant information? A crucial problem that advanced computational approaches could solve is that of specificity: how are biologically unique connections made between proteins taken from a wealth of thousands of possible partners? And conversely, how important are weak and non-specific interactions? How the binding of small ligands like drugs affects the interaction patterns and thus the function of large macromolecular aggregates? A related puzzle of paramount importance is understanding the driving forces behind the formation of amyloid-like aggregates which are associated with a number of highly-debilitating human diseases like Alzheimer's, Parkinson’s and Lou Gehrig’s diseases.
There are two major problems pertaining to predicting and designing protein-ligand interactions:
1.The free energy of complex association is difficult to estimate because both the enthalpy is not perfectly accurate, and entropy calculations is difficult to attain due to limitations in the simulation performance.
2.There is a known phenomenon, called “induced fit”, whereby a target protein undergoes conformational changes upon ligand binding. Induced fit phenomenon occurs also when two proteins associate, i.e. both of them undergo conformational changes, so that the structure of the complex is not equivalent to rigid composition of two proteins.
To address the first problem, we need scientists who develop methods for estimating changes in protein stability of binding upon mutations. To address the second problem, we need scientists who develop various docking protocols. There are specific problems of protein-protein docking, protein-peptide docking and protein-drug docking with corresponding areas of applications. This workshop can offer an outstanding platform for bringing the scientists from these fields to address these common scientific questions.
Upon binding, proteins and small-molecule ligands can undergo significant conformational changes, as schematized by the “induced fit” and “conformational selection” paradigms . Docking algorithms must therefore account for flexibility in order to allow predicting the structure of native complexes, but due to the computational cost typically only a limited flexibility is included [2,3]. In most docking schemes, first an ensemble of conformations which may be relevant for docking is obtained, e.g. from experimentally solved multiple conformers, from molecular dynamics snapshots, from normal modes analysis, essential dynamics, or rigidity theory. Then, the selected conformational space is used for docking, whether rigid or partially-flexible, and the putative complexes are ranked with scoring functions. The latter are a key ingredient and their improvement is crucial for the identification of the correct complex [4,5,6]. Progress is ongoing in the direction of predictive, efficient, and seamless treatments of both backbone and sidechain flexibility during docking and for both partners simultaneously [7,8,9].
In principle, molecular dynamics in explicit solvent can model flexibility in the most realistic way, including the possibly important role of water-mediated interactions [10,11], but the high computational cost combined with the long time-scale of conformational changes strongly limits its applicability. To attenuate this problem, enhanced-sampling techniques can be adopted [12,13,14] which allow to use more efficiently the available computer time by promptly overcoming the activation barriers. A further advantage of using molecular dynamics simulations with all-atom force fields is the possibility to access the full binding mechanism and free-energy landscape [15,16,17]. However much work is needed to obtain systematic and efficient estimates of binding free energies to compare with experiments, due both to shortcomings of available force fields and to the difficulty of exploring a large-enough portion of the vast conformational space.