GINGER - Graph Internal-coordinate Neural-network conformer Generator with Energy Refinement for Large Chemical Libraries
Conformer generation is an essential step of a variety of molecular modeling and computer-assisted drug discovery workflows such as 3D ligand-based virtual screening or fast GPU docking. GINGER (Graph Internal-coordinate Neural-network conformer Generator with Energy Refinement) is Molsoft's new cutting-edge software designed for lightning-fast high quality conformer library generation on GPUs.
GINGER utilizes state-of-the-art generative Neural Network-based algorithm to efficiently explore the conformational space of molecules, providing diverse, low-energy conformers.
To ensure high quality of conformer ensembles, raw Neural Network-generated conformers are subjected to fast gradient force-field energy minimization as needed.
End-to-end GPU acceleration: all key steps of the algorithm are implemented on GPU for maximum efficiency
Efficiency: intelligent generative algorithm achieves conformer accuracy comparable to leading industry rivals with 40% smaller ensembles
Customization Options: GINGER allows users to adjust algorithm parameters to produce conformer libraries of varying density and size based on specific research needs.
Scalability: GINGER allows efficient handling of very large molecular datasets (>1B molecules). Memory and storage-optimized implementation keeps disk/RAM hardware requirements low.
Robustness: GINGER neural network engine successfully processes >99% of compounds in a typical synthetic compound database. To further boost overall success rates to >99.9%, the remaining molecules can be seamlessly redirected for the back-up processing by our standard Monte-Carlo based sampling.
Integration and Export: output conformers can be easily exported in industry-standard formats for use with other molecular modeling software.
Generate conformers for a 10 mln compound library in one day on a single RTX4090 'gamer' workstation
Libraries >1B molecules can be processed on a small cluster or a multi-GPU server
Quickly process your in-house 2D combinatorial or AI-generated libraries, enabling application of ligand 3D structure-based methods such as field-, shape- and pharmacophore-based screening.
Apply Molsoft's field-based rapid 3D similarity search engine RIDE to your in-house libraries
Apply Molsoft's RIDGE fast structure-based GPU enabled virtual screening to your in-house libraries
MolSoft has conducted extensive testing on GINGER and a publication is expected soon. The benchmark standard for conformer generation is 'Platinum' (Friedrich et al JCIM 2017). Typically, recovery rates are reported at 0.5 / 1.0 / 1.5 Angstrom Root Mean Square Deviation (RMSD) from the bioactive conformation within an ensemble of approximately 30 conformations - considered a reasonable size for downstream applications such as RIDE and RIDGE.
Our current results demonstrate recovery rates of 45% / 80% / 93% for ensembles as small as 20 conformations on average. The median RMSD from the bioactive conformation to the closest conformation within the Ginger ensemble is 0.54 A. These results were achieved using standard GINGER settings and were validated across a benchmark dataset comprising 2859 compounds.
GINGER is a separate add-on module to MolSoft's desktop modeling software ICM-Pro and ICM-Chemist-Pro. Please contact MolSoft with any questions and to obtain an evaluation license or to purchase GINGER.