Fast, Routine Free Energy of Binding Estimation Using MovableType

Rigorous free energy of binding estimation using biomolecular structural ensembles has been of significant interest to the computational chemistry field for decades. With the advent of faster, smarter computers coupled with better algorithms and interfaces, in recent years these methods have been gaining traction in the industrial pharmaceutical space. Ultimately, the success of free energy methods rests with our ability to accurately sample the relevant conformational space (the sampling problem), while concurrently performing that sampling with the correct probability or energy (the energetics problem). The multi-dimensional integral of the partition function requires a numerical approximation for molecular ensemble sampling and for partition function estimation. Most conventional approaches to free energy methods rely on compute-intensive molecular dynamics (MD) based simulations for protein:ligand sampling. These simulations are expensive and often require specialized hardware (e.g., dedicated graphics processing units or purpose-built computer processors) utilized over hours or days for a single protein:ligand complex. This makes high-throughput screening difficult as the computational expense can become astronomical when considering more than a handful of compounds. MovableType addresses this problem by separating sampling into Local Sampling, which is treatment of protein:ligand conformer(s) using a “blurring” or “smearing” approach, and Global Sampling, which includes an ensemble of protein:ligand conformers supplied through conventional rigid-receptor/flexible-ligand docking, flexible-receptor/flexible-ligand (induced-fit) docking, rotamer/backbone (flip-state) sampling, tautomer/protomer state sampling, or even MD trajectories. In this work, we focus on results obtained with docking, and we compare these results to other free energy methods. Finally, we also touch on some new methods which will be introduced in upcoming publications.