Over the past three years, we have witnessed new challenges and opportunities in the area of drug discovery. In particular, with the rapid growth of computational resources, massively parallel and GPU computing, and the development of innovative methodology, drug discovery teams are turning from approximate to more rigorous methods. There is thus a growing demand for reliable computational approaches based on fundamental principles of statistical mechanics and trigger the need for a new CECAM workshop at the EPFL. The proposed workshop goes beyond binding affinity and focuses on the challenges in predicting the thermodynamics and kinetics of ligand binding as well as drug absorption using state-of-the-art computational approaches.
Free-energy calculations have shown promising applications in predicting protein-ligand binding. Much progress has been made in terms of efficiency and reliability by incorporating enhanced-sampling methods. For instance, 2D-replica exchange, umbrella sampling, and adaptive biasing forces have been successfully applied in PMF-based methods to calculate absolute binding free energies [1-2]. Solute tempering replica exchange has been shown to improve the results in relative binding free-energy calculations . Going beyond binding affinity, it is increasingly recognized that some compounds may exhibit comparable binding affinities but disparate enthalpic/entropic profiles, suggesting different mechanisms underlying their molecular recognition. By resolving the enthalpic/entropic endowment, a differentiation among otherwise indistinguishable compounds can be made to steer compound prioritization and optimization [4,5]. It has also been shown experimentally that some molecules binding to the same receptor can have similar binding free energies, yet different binding kinetics (koff /kon). This implies that the residence time, (1/koff), directly impacts drug efficacy and safety . Therefore, new methods are being developed to capture ligand binding/unbinding pathways to predict the underlying kinetics [7,8]. Furthermore, physics-based methods have become increasingly popular for predicting bioavailability . Rigorous free energy calculations are applied to estimate membrane permeation, active absorption through drug transporters, drug metabolism, and drug self-aggregation.
We expect to attract current experts from both academia and industry with complementary mindsets to discuss the progress achieved recently as well as persisting pitfalls. An important goal is to establish a network of international researchers to exchange and discuss methods and algorithms for the future development of computational approaches that will impact modern drug discovery.
Thrust 1: Free-energy calculations in protein-ligand binding
Binding free-energy methods based on statistical mechanics and atomistic simulations are considered to be the most rigorous route to predict ligand-binding affinities, as it can potentially capture the dynamics, desolvation, and entropic effects. Besides the well-known free-energy perturbation and thermodynamic integration methods, potential of mean force pathway method, λ-dynamics, non-equilibrium dynamics have shown promising applications in protein-ligand binding. Much progress has been made in terms of efficiency and reliability by incorporating a variety of enhanced-sampling methods. However, which method is most suited for a specific problem? What are the remaining challenges? We propose to address these questions by reaching a consensus on the best approaches that will advance binding free-energy calculations to play a major role in drug discovery.
Thrust 2: Entropy and enthalpy contributions in protein-ligand binding
To achieve high binding affinity by rational enthalpic optimization is notoriously difficult because the enthalpic gain is very often compensated by an entropic loss. The mechanism of enthalpy-entropy compensation, however, still remains a subject of controversy, and more rigorous error analysis is needed. Thus, the benefits of capturing the individual thermodynamic terms of binding are increasingly appreciated in drug discovery. However, large fluctuations of entropy in protein systems render a direct estimation of entropy via conventional simulation nearly impossible. Applying free-energy approaches to estimate the entropy cost of a protein-ligand system is still a computational intensive task. We will have a brainstorming session at the workshop on how to tackle the current entropic calculation challenge in the drug-discovery process.
Thrust 3: Receptor-ligand kinetics: kon and koff
Much less is known about the molecular determinants of binding kinetics than about those of equilibrium binding affinity, because the rate constants inherently depend on the free-energy barriers. In other words, capturing the correct ligand binding/unbinding pathways as well as the conformational fluctuations becomes essential to predict the underlying kinetics. The association rates have been estimated from simulating the spontaneous ligand-receptor association events or from multi-scale dynamics simulations. Although microsecond timescale association events are becoming accessible for atomistic simulations nowadays, the much slower dissociation rates still require novel algorithms to bridge the timescale gap. Besides the rate prediction, identifying the conformations involved in the (un-)binding pathways and the transition state will be key for the rational manipulation of binding kinetics to improve the kinetic selectivity.
Thrust 4: Bioavailability: Cell permeability, absorption, distribution, metabolism, and aggregation
Rigorous computational methods have become increasingly popular for the prediction of bioavailability as they provide atomistic insights into the permeation and distribution processes. Particular challenges in calculating the inhomogeneous diffusivity of drugs in lipid membranes and the sampling near the critical barrier region will be discussed during the workshop. Small-molecule self-aggregation has been shown to correlate with solubility and promiscuous activity of the drug. Thus, its prediction would help to prioritize compounds with desirable safety profiles and to flag unsafe compounds. We will have an overview of the current techniques and focus the discussion on the particular challenges raised recently.