The 3D LB module allows you to generate 3D molecular conformations under ligand-based constraints, leveraging a crystallized reference ligand to generate structurally and functionally similar molecules.
It operates in two steps:
First, a pre-alignment stage generates an initial conformation of the query molecule by aligning it to a reference molecule. This involves searching for common substructures with RDKit, including potentially multiple disconnected matches, and selecting the best possible atom mapping to construct the initial pose.
The second step refines the pose by optimizing the atomic coordinates under shape, pharmacophore (fully customizable), and internal-energy constraints.
In cases where an anchor is provided, the anchor atoms are enforced during pre-alignment by constraining them to appear in the common substructure search, while additional substructures may be used to align the rest of the molecule. During refinement, atoms belonging to the anchor remain fixed and are excluded from optimization.
Our unique technology demonstrated state-of-the-art performances, unlocking complex cases with large, highly flexible, or low-similarity molecules. For more information, please check the following NeurIPS paper:
Template-Guided 3D Molecular Pose Generation via Flow Matching and Differentiable Optimization
Also available as a preprint: https://arxiv.org/abs/2506.06305