To test the accuracy and precision of ATM, we applied the method to a vigorous benchmarking dataset developed by Rizzi et al.3 (https://github.com/samplchallenges/SAMPL6), where we consistently yielded a BFE estimate much closer to experimental values than those displayed by established methods that utilized similar or greater computational expense. One of the sample host-guest systems that we tested on, as shown in Figure 1 above, consisted of the host, cucurbituril (CB8), and its guest, quinine (G3). The other two complexes (not shown) consisted of an octa-acid (OA) host and its two guests, 5-hexenoic acid (G3) and 4-methylpentanoic acid (G6).
Experimental data collected for the host-guest complexes yielded BFE of -5.18 ± 0.02 kcal/mol for OA-G3, -4.97 ± 0.02 for OA-G6, and -6.45 ± 0.06 kcal/mol for CB8-G3. Well-converged computational results, on average, overestimated (more negatively) BFE estimates by -1.2 kcal/mol, -2.1 kcal/mol, and -4.4 kcal/mol, respectively. In comparison, ATM predicted BFE estimates of -5.89 ± 0.33 kcal/mol, -6.32 ± 0.21 kcal/mol, and -8.53 ± 0.64 kcal/mol, respectively; which differed by -0.71 kcal/mol, -1.35 kcal/mol, and -2.08 kcal/mol from the experimental values. ATM, compared to the other well-converged computational methods, had a significantly closer estimate with respect to the experimental values. Equilibration analysis of each complex’s five conformations also converged at 5 ns equilibration time with BFE estimates consistent with ATM.
Considering the relatively low computational cost compared to other methods, ATM’s statistical uncertainties indicate that comparable levels of reproducibility and computational efficiency could be achieved. As such, we would like to validate ATM’s method and implementation on the SAMPL6 benchmark host-guest dataset.
While ATM is an alchemical method, its implementation is very straightforward and achieves results akin to physical pathway methodologies. ATM does not require splitting the alchemical transformations in electrostatic and non-electrostatic steps, alchemical topologies, and soft-core pair potentials, and is potential-function agnostic, with functionality that can be extended to other more advanced many-body potentials. Additionally, ATM does not require any modification of the OpenMM core energy routines, and is freely available via a plugin of the OpenMM molecular simulation package.