Automated Coarse-Grained Mapping Algorithm for the Martini Force Field and Benchmarks for Membrane-Water Partitioning

Abstract

With a view to high-throughput simulations, we present an automated system for mapping and parameterizing organic molecules for use with the coarse-grained Martini force field. The method scales to larger molecules and a broader chemical space than existing schemes. The core of the mapping process is a graph-based analysis of the molecule's bonding network, which has the advantages of being fast, general, and preserving symmetry. The parameterization process pays special attention to coarse-grained beads in aromatic rings. It also includes a method for building efficient and stable frameworks of constraints for molecules with structural rigidity. The performance of the method is tested on a diverse set of 87 neutral organic molecules and the ability of the resulting models to capture octanol-water and membrane-water partition coefficients. In the latter case, we introduce an adaptive method for extracting partition coefficients from free-energy profiles to take into account the interfacial region of the membrane. We also use the models to probe the response of membrane-water partitioning to the cholesterol content of the membrane.

Figure 1

Flowchart outlining the three main stages of the algorithm: mapping of atoms onto beads and parameterization of nonbonded and bonded interactions. In the mapping and bonded interaction stages, the procedure differs for ring and nonring fragments, while for nonbonded interactions, the steps for aliphatic and aromatic fragments are distinct.

Figure 2

(a) Predefined mappings for rings with examples where these fragments occur. The fragments involved are in bold, with red ellipses indicating the coarse-grained beads. (b) The iterative mapping process for a triphenylene ring fragment.

Figure 3

Mapping scheme for trihexylamine, including two iterations of graph-based grouping followed by postprocessing to remove a single-atom bead.

Figure 4

Procedure for constructing full rings from mapped ring fragments.

Figure 5

Schematic of the geometric constructions used to determine weights for virtual sites when (a) n = 4 and (b) n = 3.

Figure 6

Schematic illustration of the process for generating virtual sites and ring constraints for warfarin. (a) Mapping of atoms onto coarse-grained beads (red ellipses), (b) ring constraints (thick lines), real sites (filled circles), and a virtual site (open circle), (c) dihedral constraint for the double ring (dashed red line), and (d) bonds connecting the ring fragments (thick dashed lines).

Figure 7

Octanol–water partition coefficients (log K_OW) for a set of organic molecules using coarse-grained models compared to experiment (dot markers) or predicted ALOGPS values (cross markers) where experimental data are not available.

Figure 8

Octanol–water partition coefficients (log K_OW) for amino acids calculated using Martini 2.2 and automated models compared to experimental values.

Figure 9

(a) Convergence of log K_MW with cutoff between the membrane and water for ethylene glycol and trihexylamine. (b) Free-energy profiles for the two compounds, with cutoff from an RMSD threshold of 0.1 indicated by crosses. The center of the membrane lies at r = 0.

Figure 10

Membrane–water partition coefficients (log K_MW) for a range of small organic molecules using coarse-grained models. (a) Coarse-grained vs experimental in a pure phospholipid membrane and (b) log K_MW in a phospholipid + 30% cholesterol vs pure phospholipid membrane, both from the coarse-grained models.

Figure 11

Potentials of mean force between the bilayer center (r = 0.0) and bulk water (r = 5.0) for (a) tetrachlorocatechol and (b) trihexylamine, in POPC and POPC + 30% cholesterol.

Figure 12

Membrane–water partition coefficients (log K_MW) for alkyl sulfates (chain lengths 8, 10, 12, and 13) and sulfonates (chain lengths 8, 10, and 12–14), with models generated with predefined fragment matching (solid lines) and with the standard mapping procedure (dashed lines). In all cases, increasing the chain length increases log K_MW.

Figure 13

CPU time of the parameterization for linear alkanes of different chain lengths, with and without UFF minimization of the 200 atomistic conformers. Timings for automartini are also included for reference.