Search papers, labs, and topics across Lattice.
This study introduces DBMol, a novel framework for designing high-affinity small molecules by leveraging advanced structure prediction models like AlphaFold-3 and Boltz-2. By employing an alternating optimization and projection process, DBMol enhances pocket-specific interactions and binding affinity while ensuring the generation of chemically valid molecules. Experimental results demonstrate that DBMol significantly improves pocket coverage and maintains molecular diversity, outperforming traditional unconditional generation methods and showing competitive performance on held-out metrics without reference-ligand supervision.
DBMol transforms small molecule design by effectively optimizing binding affinities while ensuring chemical validity, setting a new standard for de novo molecular generation.
Designing small molecule ligands that bind with high affinity to specific protein pockets is a fundamental goal in drug discovery, as small molecules constitute a major fraction of approved therapeutics. Recent breakthroughs in structure prediction, such as AlphaFold-3 and Boltz-2, enable accurate biomolecular interaction prediction and show promise as foundation models for downstream tasks, including binding affinity prediction. We propose to leverage these models and introduce DBMol, a new structure predictor-guided framework for de novo small molecule design. DBMol formulates an alternating optimization and projection process. In the optimization stage, DBMol starts from an initial molecule and uses gradient-based optimization to improve pocket-specific interactions and predicted binding affinity using a structure prediction model. In the projection stage, a flow-matching model maps the optimized molecular graph to discrete and chemically valid molecules. Experiments show that DBMol effectively optimizes the Boltz-2 affinity proxy and generates molecules with strong predicted affinity and specificity under Boltz-2 evaluation. To reduce self-confirmation bias, we further evaluate generated molecules using held-out metrics, including AF3-based evaluation. DBMol substantially improves pocket coverage while maintaining molecular diversity over unconditional generation, and is competitive under held-out metrics despite the absence of reference-ligand supervision. These results support the promise of structure prediction models as effective optimization signals for de novo molecular design.
