CombiRIDGE

CombiRIDGE is a GPU-accelerated, high-throughput combinatorial docking and virtual screening platform developed by Molsoft L.L.C. It is designed for drug discovery teams that need to efficiently explore ultra-large or combinatorial chemical libraries, enabling the screening of billions to trillions of compounds against a target binding site. The platform integrates three core technologies: GINGER, a generative neural network conformer enumeration method; RIDGE, an ultra-fast GPU docking engine; and RTCNN, an advanced graph neural network scoring function.

CombiRIDGE employs an anchor-driven approach that begins with a 3D core fragment anchored within the binding site to preserve critical molecular interactions. From this starting point, full 3D conformers are generated directly at the binding site using GINGER, enabling rapid combinatorial expansion of pre-docked anchor structures into complete compounds that comprehensively sample the library space around each anchor fragment. Pose refinement is performed via GPU-accelerated grid/MMFF energy minimization, and final candidate evaluation is carried out using the RTCNN scoring function to identify high-quality hits.

Key Capabilities

• GPU-accelerated docking and screening powered by the RIDGE engine, enabling ultra-high throughput processing on a single GPU
• Generative neural network conformer generation via GINGER, producing 3D conformers directly at the binding site
• Advanced graph neural network scoring with RTCNN for robust evaluation of docked poses
• Anchor-driven combinatorial expansion that maintains critical binding site interactions while exploring vast chemical spaces
• GPU-accelerated grid/MMFF energy minimization for pose refinement and accuracy
• Support for ultra-large and combinatorial library screening spanning billions to trillions of compounds

Supported Anchor Libraries

• Enamine REAL: approximately 45 million anchor molecules covering a total chemical space of approximately 48 billion compounds
• Wuxi Galaxi: approximately 35 million anchor molecules covering a total chemical space of approximately 15 billion compounds
• eMolecules: approximately 60 million anchor molecules covering a total chemical space of over 1.2 trillion compounds, with more than 7 trillion expected soon
• Additional vendor libraries are planned for future inclusion

Demonstrated Performance: 28 Billion Enamine REAL Compounds

1. Anchor pre-processing with GINGER: approximately 18 million anchor molecules converted to 3D conformers at a rate of 7,500 per minute, completed within 40 hours on a single Nvidia RTX 4090 GPU
2. Anchor docking with RIDGE: small, low-rotatability anchor molecules docked at 3,500 per minute over approximately 90 hours, identifying poses with exposed expansion sites
3. Top anchor expansion using in-situ GINGER combined with GPU-accelerated grid/MMFF minimization: completed in just 6 hours

Validation: Venetoclax Re-Docking Test Case

• Venetoclax, a BCL-2 inhibitor with an unusually high molecular weight of 868.44 g/mol, was selected as a challenging benchmark to test CombiRIDGE's ability to rebuild and re-dock a complex molecule
• A set of 1 million anchors from the Enamine REAL library and a hypothetical Venetoclax synthon were docked and scored
• Following CombiRIDGE enumeration using GINGER and energy minimization, the re-docked pose achieved an RMSD of 1.07 Å relative to the crystal structure, demonstrating high accuracy for a large, complex molecule

CombiRIDGE is optimized to run on modern Nvidia GPU hardware, with demonstrated performance on a single RTX 4090. The platform integrates seamlessly with curated anchor libraries from leading chemical vendors and is positioned as a scalable solution for structure-based virtual screening campaigns targeting vast combinatorial and ultra-large chemical spaces.