Final Report Summary - SWARMDESIGNTOOLS (Swarm Intelligence Simulations as Tools for Molecular Design of Better Medicines)
In this project, we have introduced a new method for computation of molecular free energies. Free energy is the key quantity governing change in chemistry and biology, from phase transitions to interactions between proteins and small molecules. Indeed, a new generation of potent, selective and safe medicines is required to drive advances in healthcare. In the pharmaceutical industry, the design of novel medicines is typically based on optimising the interaction of a small molecular ligand with its protein target.
Although computational tools are widely used within the industry to guide small molecule design, it is widely acknowledged that these computational approaches lack sufficient accuracy for reliable application. The most accurate computational tools to date are based on statistical mechanics, and usually employ molecular dynamics (MD) or Monte Carlo simulations to estimate the free energy change of a molecular system, corresponding to for example a chemical change or intermolecular association. These methods however struggle to account for the full range of structures that contribute to the change, leading to only a semi-quantitative estimate of free energy.
To address this, we have introduced a new molecular dynamics simulation approach inspired by the cooperative behaviour of birds swarming and fish schooling. This swarm-enhanced sampling molecular dynamics (sesMD) algorithm couples together multiple simulation copies of the molecular system as van der Waals-like particles. The attractive and repulsive components of their interaction enable a cohesive behaviour which leads to sampling of low and higher energy molecular conformations. We have demonstrated broader sampling of structures for model peptide and hydrocarbon systems, and also for a protein kinase. For the latter, a commonly-targetted class of protein in structure-based drug design, we showed that sesMD was capable of sampling a wide range of conformations, corresponding to a variety of different experimentally-determined crystal structures. These structures were not sampled by multiple independent MD simulations.
The main focus of this project has been in the implementation of sesMD into a free energy framework. We have combined sesMD with the thermodynamic integration (TI) methodology to produce the sesTI method. TI links two thermodynamic states via a coupling parameter,λ, and gathers the integral over dG/dλ as the system is stepwise mutated from its initial to final state. Our sesTI algorithm efficiently utilises a parallel architecture to accumulate this free energy over the full coupled swarm of simulation replicas.
We find that the improved sampling of higher energy structures leads to a quantitative estimate of free energy for the key benchmark of the butane-to-butane transformation in water (this type of testcase is often used as the free energy change is independent of simulation conditions or molecular model employed). The estimate is superior to the free energy estimate based on the TI average of many independent MD simulations. Whilst further testing of sesTI on a range of different chemical transformations is ongoing, the performance of the approach for the exacting butane testcase points to its utility in quantifying free energy changes. Indeed, both the sesMD and sesTI methods hold considerable promise as useful new software tools for molecular engineering. These approaches can guide the design of new therapeutics within the pharmaceutical industry but also provide new insights into the factors governing recognition between molecules and other important chemical and biological phenomena at thermodynamic equilibrium.
Although computational tools are widely used within the industry to guide small molecule design, it is widely acknowledged that these computational approaches lack sufficient accuracy for reliable application. The most accurate computational tools to date are based on statistical mechanics, and usually employ molecular dynamics (MD) or Monte Carlo simulations to estimate the free energy change of a molecular system, corresponding to for example a chemical change or intermolecular association. These methods however struggle to account for the full range of structures that contribute to the change, leading to only a semi-quantitative estimate of free energy.
To address this, we have introduced a new molecular dynamics simulation approach inspired by the cooperative behaviour of birds swarming and fish schooling. This swarm-enhanced sampling molecular dynamics (sesMD) algorithm couples together multiple simulation copies of the molecular system as van der Waals-like particles. The attractive and repulsive components of their interaction enable a cohesive behaviour which leads to sampling of low and higher energy molecular conformations. We have demonstrated broader sampling of structures for model peptide and hydrocarbon systems, and also for a protein kinase. For the latter, a commonly-targetted class of protein in structure-based drug design, we showed that sesMD was capable of sampling a wide range of conformations, corresponding to a variety of different experimentally-determined crystal structures. These structures were not sampled by multiple independent MD simulations.
The main focus of this project has been in the implementation of sesMD into a free energy framework. We have combined sesMD with the thermodynamic integration (TI) methodology to produce the sesTI method. TI links two thermodynamic states via a coupling parameter,λ, and gathers the integral over dG/dλ as the system is stepwise mutated from its initial to final state. Our sesTI algorithm efficiently utilises a parallel architecture to accumulate this free energy over the full coupled swarm of simulation replicas.
We find that the improved sampling of higher energy structures leads to a quantitative estimate of free energy for the key benchmark of the butane-to-butane transformation in water (this type of testcase is often used as the free energy change is independent of simulation conditions or molecular model employed). The estimate is superior to the free energy estimate based on the TI average of many independent MD simulations. Whilst further testing of sesTI on a range of different chemical transformations is ongoing, the performance of the approach for the exacting butane testcase points to its utility in quantifying free energy changes. Indeed, both the sesMD and sesTI methods hold considerable promise as useful new software tools for molecular engineering. These approaches can guide the design of new therapeutics within the pharmaceutical industry but also provide new insights into the factors governing recognition between molecules and other important chemical and biological phenomena at thermodynamic equilibrium.