Molecular diffusion improvements boost drug design tasks without model retraining
Budgeted Quotient-Residual Guidance for Frozen Pocket-Conditioned Molecular Diffusion
Machine Learning
Summary
Many methods that generate molecules change their shapes but don't always match what drug designers need, like specific distances or contacts within the molecule. The authors developed a way called budgeted quotient-residual guidance (QRG) that adjusts molecule generation steps on the fly to better fit these design goals without redoing the entire model. Their approach improves the quality of generated drug-like molecules in several test tasks, keeping variety and newness high. This method works with existing molecular diffusion tools and speeds up achieving better results.
What this means in practice
- •For drug discovery teams: Enhance molecule generation quality in fragment growing, scaffold hopping, and linker design tasks using existing diffusion models without retraining.
- •For chemical informatics engineers: Integrate runtime corrections that respect molecular structural goals into frozen molecular samplers to improve ligand generation workflows.
Authors
Xinyu Wang, Jinbo Bi, Minghu Song
Abstract
Pocket-conditioned molecular diffusion updates ambient atom coordinates, but many lead-optimization objectives are expressed on quotient features such as distances, contacts, and anchored substructures. We introduce budgeted quotient-residual guidance (QRG), an inference-time correction that makes these quotient objectives active without retraining the molecular generator. QRG lifts quotient covectors to metric-horizontal ambient directions and delivers them through a trust budget set by the frozen sampler's own step norm: quotient geometry chooses the direction, while sampler motion bounds the scale. We derive the horizontal lift, closed-form sampler-budget update, KL/kinetic interpretation around a frozen reverse step, equivariance conditions, and a product-budget split for budget-capped section and residual controls. Controlled quotient tasks confirm that sampler-relative delivery activates signals that raw local quotient gradients leave dormant. On frozen TargetDiff backbones, official seed-0 CBGBench ligand-generation/editing sweeps show practical quality-runtime gains: Local-QRG improves validity from 0.815 to 0.864 on fragment growing, 0.664 to 0.707 on scaffold hopping, and 0.681 to 0.712 on linker design, while PredNext-QRG improves fragment/scaffold and remains near-neutral on linker. Novelty remains 1.000 and diversity is preserved in the matched multi-seed molecular slice, giving task-dependent improvements without sampler retraining or backbone modification. Overall, QRG provides a lightweight route to quotient-aware inference for frozen molecular samplers with explicit runtime accounting.