GINGER

GINGER (Graph Internal-coordinate Neural-network conformer Generator with Energy Refinement) is a cutting-edge conformer generation software developed by Molsoft, designed for lightning-fast, high-quality conformer library generation on GPUs. It is purpose-built for computational chemists and drug discovery teams who need to process large chemical libraries — from millions to over a billion compounds — as part of workflows such as 3D ligand-based virtual screening, field- and shape-based screening, pharmacophore-based screening, and fast GPU docking.

GINGER employs a state-of-the-art generative neural network algorithm based on a directional graph convolution architecture operating in internal coordinates, enabling efficient exploration of molecular conformational space. The approach was published by Eugene Raush, Ruben Abagyan, and Maxim Totrov in the Journal of Chemical Theory and Computation (2024).

Key Features

• Neural network-driven conformer generation: GINGER uses a generative graph convolution neural network to efficiently explore conformational space and produce diverse, low-energy conformers.
• Energy refinement: Raw neural network-generated conformers are subjected to fast gradient force-field energy minimization to ensure high-quality ensembles.
• End-to-end GPU acceleration: All key algorithmic steps are implemented on GPU for maximum computational efficiency.
• Smaller, accurate ensembles: The intelligent generative algorithm achieves conformer accuracy comparable to leading industry tools while producing ensembles approximately 40% smaller.
• Customization: Users can adjust algorithm parameters to control the density and size of conformer libraries to suit specific research requirements.
• Scalability: Designed to handle very large molecular datasets exceeding 1 billion molecules, with memory- and storage-optimized implementation to keep hardware requirements low.
• Robustness: The neural network engine successfully processes over 99% of compounds in a typical synthetic compound database. For the remaining molecules, seamless fallback to a Monte-Carlo-based sampling method boosts overall success rates to over 99.9%.
• Export and integration: Output conformers can be exported in industry-standard formats for use with other molecular modeling software.

Performance and Benchmarks

• Benchmarked against the 'Platinum' standard (Friedrich et al., JCIM 2017) across a dataset of 2,859 compounds.
• Achieves recovery rates of 45% / 80% / 93% at 0.5 / 1.0 / 1.5 Å RMSD from the bioactive conformation using ensembles as small as 20 conformations on average.
• Median RMSD from the bioactive conformation to the closest conformation in the GINGER ensemble is 0.54 Å.
• A 10 million compound library can be processed in a single day on one RTX 4090 consumer-grade GPU workstation.
• Libraries exceeding 1 billion molecules can be processed on a small cluster or a multi-GPU server.

Workflow and Integration Benefits

• Rapidly convert in-house 2D combinatorial or AI-generated libraries into 3D conformer ensembles, enabling downstream 3D structure-based methods.
• Directly compatible with Molsoft's RIDE field-based rapid 3D similarity search engine for in-house library screening.
• Directly compatible with Molsoft's RIDGE fast structure-based GPU-enabled virtual screening platform.
• Supports processing of both standard synthetic compound databases and novel AI-generated chemical libraries.

GINGER is available as a separate add-on module for Molsoft's desktop modeling software ICM-Pro and ICM-Chemist-Pro, or can be run as a service by Molsoft. Evaluation licenses and purchase options are available directly through Molsoft.