Molecules with rotatable bonds can exist in multiple conformations, and locating the lowest energy conformation is crucial for ensuring the accuracy of subsequent tasks. Running a conformational search can also reveal the conformational preferences of different bonds, and show how strained various alternate conformations are relative to the global minimum.
Rowan's conformational searching workflow proceeds in three stages:
An example API script is available on GitHub.
By default, Rowan generates conformers using openconf. openconf combines ETKDG seed structures with torsional multiple-minimum Monte Carlo exploration. Trial structures are locally minimized, filtered by energy, and geometrically deduplicated before a representative ensemble is selected. The default search performs 200 Monte Carlo steps, keeps candidates within 10 kcal/mol of the lowest-energy structure, samples parent conformers with softmax weighting, and selects a diverse final ensemble.
Rowan supports a variety of conformer generation/search methods:
| Conformer generation/search method | Accepts 2D inputs? | Accepts 3D inputs? | Blog post |
|---|---|---|---|
| openconf (default) | Yes | Yes | openconf newsletter |
| ETKDG (through RDKit) | Yes | Yes | |
| Multiple-minimum Monte Carlo (MCMM or MMMC) | No | Yes | MMMC blog post |
| iMTD-GC (through CREST) | No | Yes |
Each method has different settings associated with it. These alternatives are useful when reproducing an older protocol or when a molecule benefits from a particular sampling strategy. For more information about these methods, read our blog post "How to Find Conformers."
In the API, conformer generation/search can be configured with conf_gen_settings. This step can be skipped by passing initial_conformers, as in this example script.
To reduce the number of conformers that need to be optimized and thereby reduce overall runtime, conformer clustering can be requested. Rowan uses k-means clustering to group conformers by three-dimensional shape and surface properties, then selects representative structures from each cluster for optimization rather than optimizing the full generated ensemble. This loosely follows the approach of Wilcken and co-workers.
Clustering behavior can be tuned along the following parameters:
num_clusters, default 5)conformers_per_cluster, default 3)By default, clustering uses the following six descriptors:
solvent_accessible_surface_area)polar_solvent_accessible_surface_area)radius_of_gyration)plane_of_best_fit)normalized_principal_moment_ratio_1)normalized_principal_moment_ratio_2)In the API, clustering can be configured with conformer_clustering_settings.
The generated ensemble is optimized and scored using the user's multistage optimization setting , with solvent corrections applied when requested. Keeping generation separate from optimization makes it possible to use inexpensive sampling while reserving a more accurate method for the final geometries and energies.
In the API, optimization can be configured with either final_method and solvent or multistage_opt_settings.
Rowan's conformational searching workflow supports transition-state (TS) conformer searches. A transition state (TS) conformer search should be performed on fully optimized TS, not a guess TS.
Transition states require extra care, as conformer generation schemes will natively relax the TS to a local minimum. User provided constraints are required to ensure that the conformer generation creates conformers that look like the TS. Once the conformers have been generated, further TS optimizations are performed without constraints.