Conformer Generation for Structure-Based Drug Design: How Many and How Good?

Abstract

Conformer generation, the assignment of realistic 3D coordinates to a small molecule, is fundamental to structure-based drug design. Conformational ensembles are required for rigid-body matching algorithms, such as shape-based or pharmacophore approaches, and even methods that treat the ligand flexibly, such as docking, are dependent on the quality of the provided conformations due to not sampling all degrees of freedom (e.g., only sampling torsions). Here, we empirically elucidate some general principles about the size, diversity, and quality of the conformational ensembles needed to get the best performance in common structure-based drug discovery tasks. In many cases, our findings may parallel “common knowledge” well-known to practitioners of the field. Nonetheless, we feel that it is valuable to quantify these conformational effects while reproducing and expanding upon previous studies. Specifically, we investigate the performance of a state-of-the-art generative deep learning approach versus a more classical geometry-based approach, the effect of energy minimization as a postprocessing step, the effect of ensemble size (maximum number of conformers), and construction (filtering by root-mean-square deviation for diversity) and how these choices influence the ability to recapitulate bioactive conformations and perform pharmacophore screening and molecular docking.

Systematic Comparison of Experimental Crystallographic Geometries and Gas-Phase Computed Conformers for Torsion Preferences

“…As noted above, while CREST attempts to generate exhaustive ensembles under 6 kcal/mol from the global minima, in general, only a few conformers are generated (Figures S5 and S6), and 250 conformers is sufficient to encompass 93% of the COD set and 85% of the Platinum Diverse set (Figure S5). While larger ensembles tend to yield smaller RMSD, a comparison is still useful despite some differences in ensemble size …”

“…Finally, we note that while most previous work has compared computed conformers with experimental crystal structures or bound-ligand geometries, recent work has used cross-sectional areas , and rotational spectroscopy − to gain geometric insight into gas-phase conformer geometries. Future work should consider such alternate benchmarks since bound conformations, in particular, reflect stabilization from intermolecular interactions …”

“…We note somewhat worse performance across the Platinum set than that of ETKDG, particularly for molecules with more rotatable bonds. There is a notable difference in the radius of gyration between the Platinum and low-energy CREST/GFN2 compounds, particularly with increasing numbers of rotatable bonds, likely indicating some preference for extended conformations in the Platinum set due to intermolecular interactions of bound ligands, not present in an isolated gas-phase calculation …”

STORMM: Structure and TOpology Replica Molecular Mechanics for chemical simulations

“…In many situations, it is sufficient to accumulate quantities such as the representation of a force with a fixed-precision scaling of perhaps 2 24 , such that most accumulations will not go outside the range [−2 31 , 2 31 ). However, if an accumulation does stray from this range, even on occasion, the result can be catastrophic: in the case of forces, a large negative force flips to become a large positive force.…”

“…However, for the professional-grade A100 GPUs the performance of 64-bit accumulation in shared resources appears to be as fast as solitary 32-bit accumulation, supplanting split accumulation by a small amount. On all cards, the cost of each atomic reaches a plateau as the accumulations approach 2 31 , the maximum amount that the primary accumulator can accept in a single operation and also the most likely to trigger incremental overflow. The split accumulation carries an advantage over direct 64-bit accumulation in global memory when the secondary accumulators reside there (atomic operations to global memory addresses take place in the global L2 cache), and even, in the case of the A40 GPU, when both accumulators reside there.…”