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, which support the promise of structure prediction models as effective optimization signals for de novo molecular design.
Abstract
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.
MolecularCanvas is an interactive system that enables users to iteratively construct an optimization context by integrating high-level goals, structure-level annotations, property constraints, and reference-based preferences that guides the generation of candidate molecules across diverse molecular structures.
Haoyu Dong, Rui Sheng, Shu-Hao Zhang et al.· 0 citations
DyAb is a pair-wise representation built on top of a pre-trained protein language model that achieves a Spearman rank correlation of up to 0.85 on binding affinity prediction across monoclonal antibodies targeting three different antigens.
J. Lin, Jennifer L. Hofmann, Andrew Leaver-Fay et al.· mAbs· 0 citations
A novel method named Perturbed Flow Matching (PFM), which significantly reduces sampling steps by leveraging a Flow Matching framework and introduces a unique perturbed conditional probability path design that incorporates pocket binding site information and atom type-coordinate coupled information to enhance molecular generation performance.
A clear pattern is revealed in LLM spatial capabilities: while they still lag behind state-of-the-art approaches, they are promising and can handle multiple spatial constraints simultaneously, enabling scaling to heterogeneous setups.
Thomas MacDougall, Maksim Kuznetsov, Roman Schutski et al.· arXiv.org· 1 citation
Plant-derived small molecules possess highly diverse physicochemical properties, and the computational design of their protein recognition elements depends not only on the global structural quality of candidate backbones, but also on whether the local binding pocket, ligand-contact pattern, and predefined recognition conformation can be consistently retained after sequence design and structural back-prediction. To explore pocket-design strategies for different types of natural-product small molecules, this study selected capsaicin, (4R)-limonene, and quercetin as model ligands, representing a flexible amphipathic molecule, a compact hydrophobic monoterpene, and a rigid polyphenolic flavonoid scaffold, respectively, and covering the dimensions of pungent sensory flavor, volatile aroma, and flavonoid functional constituents. A ligand- physicochemical-property-guided computational design and multi-stage prioritization framework was established for candidate protein binders. The results showed that candidates with favorable initial global structural scores did not necessarily form reasonable local small-molecule binding pockets, indicating that evaluation of the local ligand environment is essential for candidate prioritization. After screening, 31 partial- pocket candidate backbones for capsaicin, 75 buried hydrophobic-pocket candidate backbones for (4R)-limonene, and 56 pocket-qualified candidate backbones for quercetin were obtained. Further sequence design and structural back-prediction analyses indicated that a subset of candidates could maintain the original pocket geometry and major ligand-contact patterns after sequence realization. Overall, these results suggest that the physicochemical properties of different plant-derived small molecules substantially influence the efficiency of de novo protein pocket formation, with compact hydrophobic ligands being more compatible with buried hydrophobic- pocket strategies, whereas flexible or multipolar ligands require a more refined balance between hydrophobic burial and polar exposure. This study provides a pre- experimental computational prioritization framework for natural-product small- molecule-recognizing proteins and offers candidate resources for subsequent protein expression, in vitro binding validation, active-constituent enrichment, and development of small-molecule biorecognition tools. Graphical Abstract
Biomolecular interactions, including protein–protein interactions, protein–nucleic acid recognition, and protein–small molecule binding, underlie a wide range of biological processes and therapeutic mechanisms. Although recent de novo design methods can generate candidate binders for diverse molecular targets, practical design campaigns remain limited by low filter-passing rates and model-specific biases that arise when designs are optimized against a single predictor. Here, we present RFOptimization (RFO), a training-free framework for all-atom biomolecular binder optimization. RFO formulates binder improvement as a residue-wise mutational search problem, sampling candidate substitutions alternately based on gradient-guided sequence optimization using all-atom structure prediction models and a cycling-based sequence redesign strategy that alternates structure generation with an orthogonal predictor and MPNN-based sequence design to improve the in silico success rate of RFdiffusion-generated binders within minutes of computation. To reduce overfitting to any individual structure model, candidate mutations are further evaluated with orthogonal AlphaFold3 metrics as final filters. We demonstrate the generality of RFO across diverse design settings, including classical protein binder design, ligand-binding biosensor design, cyclic peptide design, and active site-aware enzyme design.
Unknown authors· bioRxiv· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
