RIDE: Reference-Anchored Inference-Time Diffusion Editing for Scaffold Hopping

Ruoxi Gao; Frazier N. Baker; Trieu Nguyen; Xia Ning

cs.AISep 28, 2026

RIDE: Reference-Anchored Inference-Time Diffusion Editing for Scaffold Hopping

Authors: Ruoxi Gao, Frazier N. Baker, Trieu Nguyen, Xia Ning

Organizations: Department of Computer Science and Engineering, The Ohio State University · Department of Biomedical Informatics, The Ohio State University · Translational Data Analytics Institute, The Ohio State University · Division of Medicinal Chemistry and Pharmacognosy, The Ohio State University

Abstract

Scaffold hopping is a critical task in drug discovery, which seeks to discover new, structurally distinct molecules that share key functional groups and similar 3D shape with a reference binding ligand. Existing diffusion-based scaffold hopping methods formulate the problem as conditional generation of scaffolds given the functional groups. However, they lack a principled mechanism to jointly enforce 2D structural novelty and preserve the 3D shape of the reference ligand. Here, we introduce RIDE, a Reference-anchored Inference-time Diffusion Editing framework for scaffold hopping. RIDE recovers the reference diffusion noise trajectory conditioned on the binding pocket and functional groups, selects an optimal trajectory segment for editing via noise perturbation, and conducts a value-guided scaffold sampling to generate new scaffolds. Extensive experimental results demonstrate that, compared to baselines, RIDE consistently generates scaffolds with lower 2D similarity and higher 3D similarity to the reference, with an average improvements of 11.7% and 7.3%, respectively. Further analysis reveals that RIDE can accommodate various reward functions, and can preserve 3D similarity even when this is not explicitly included in the reward. Two case studies illustrate RIDE's ability to generate distinct scaffolds with different structures and properties, and its ability to introduce substantial 2D variation while maintaining very high 3D similarity. RIDE is publicly available at https://anonymous.4open.science/r/RIDE-C8A0.

Figures & tables

Figure 1: Overview of $\mathtt{RIDE}$ . a , $\mathtt{RIDE}$ generates new scaffolds based on a reference scaffold, fixed functional groups, and protein pocket; b , $\mathtt{RIDE}$ recovers the noise trajectory for the reference ligand from a diffusion-based molecule generation model adapted to scaffold hopping; c , $\mathtt{RIDE}$ identifies a reference-optimal noise trajectory segment for editing; and d , $\mathtt{RIDE}$ performs a value-guided transition within the reference-optimal trajectory segment for a controlled editing of the noise trajectory, thus sampling a set of new scaffolds.

	Model	Similarity		Vina			QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
	Model	Sim ${}_{\text{2D}}$ $\downarrow$	Sim ${}_{\text{3D}}$ $\uparrow$	Vina S $\downarrow$	Vina M $\downarrow$	Vina D $\downarrow$	QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
Baselines	conDitar-a	$0.398$	$0.795$	$-7.192$	$-7.753$	$-8.564$	$0.455$	$0.622$	$85.2$
	IPDiff-a	$0.420$	$0.856$	$\mathbf{-7.706}$	$\mathbf{-8.176}$	$\mathbf{-8.897}$	$0.456$	$0.619$	$94.7$
	DiffHopp	$0.426$	$0.812$	$-2.223$	$-6.209$	$-8.642$	$\mathbf{0.535}$	$0.667$	$89.5$
	ShEPhERD	$0.417$	$0.881$	$-6.471$	$-7.402$	$-8.390$	$0.404$	$0.576$	$29.4$
$\mathtt{RIDE}^{\text{(1)}}$	conDitar-a	$0.354$	$0.884$	$-7.508$	$-8.049$	$\underline{-8.833}$	$0.427$	$0.601$	$91.5$

Table 1: Comparison of

\mathtt{RIDE}

with baselines on the test set with

L=100

. Best and second-best results are shown in bold and underlined , respectively.

	$L$		Similarity		Vina			QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
	$L$		Sim ${}_{\text{2D}}$ $\downarrow$	Sim ${}_{\text{3D}}$ $\uparrow$	Vina S $\downarrow$	Vina M $\downarrow$	Vina D $\downarrow$	QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
random perturbation (Eq. 11 )	$T/10$	$\mathtt{RIDE}^{\text{(1)}}$	$0.387$	$\mathbf{0.898}$	$-7.742$	$-8.213$	$-8.840$	$0.442$	$0.614$	$86.6$
	$T/10$	$\mathtt{RIDE}^{\text{(2)}}$	$0.343$	$0.874$	$-7.581$	$-8.091$	$-8.813$	$0.425$	$0.604$	$90.0$
	$T$	$\mathtt{RIDE}^{\text{(1)}}$	$0.391$	$0.866$	$\mathbf{-7.837}$	$\mathbf{-8.308}$	$\underline{-9.028}$	$\mathbf{0.462}$	$\underline{0.629}$	$92.3$
	$T$	$\mathtt{RIDE}^{\text{(2)}}$	$0.356$	$0.854$	$-7.763$	$-8.232$	$-8.915$	$0.450$	$0.620$	$95.2$
value-guided sampling (Eq. 18 )	$T/10$	$\mathtt{RIDE}^{\text{(1)}}$	$0.354$	$0.884$	$-7.508$	$-8.049$	$-8.833$	$0.427$	$0.601$	$91.5$

Table 2: Comparison of random perturbation and value-guided sampling on

L=T/10

and

L=T

(perturbation segment starting from

t_{1}^{\ast}

to the end of the diffusion trajectory), using conDitar-a as the base model. Best and second-best results are shown in bold and underlined , respectively.

Figure 2: Comparison of Sim ${}_{\text{2D}}$ vs. Sim ${}_{\text{3D}}$ across $t_{1}\in\{nL\}_{n=5}^{9}$ and $t_{1}^{\ast}$ with $L=T/10$ . (a) conDitar-a; (b) IPDiff-a.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Inversion	Sampling		Similarity		Vina			QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
Inversion			Sim ${}_{\text{2D}}$ $\downarrow$	Sim ${}_{\text{3D}}$ $\uparrow$	Vina S $\downarrow$	Vina M $\downarrow$	Vina D $\downarrow$	QED $\uparrow$	SA $\uparrow$	Conn. (%) $\uparrow$
✓	random perturbation	$\mathtt{RIDE}^{\text{(1)}}$	$0.391$	$0.866$	$\mathbf{-7.837}$	$\mathbf{-8.308}$	$\underline{-9.028}$	$\mathbf{0.462}$	$\underline{0.629}$	$92.3$
		$\mathtt{RIDE}^{\text{(2)}}$	$0.356$	$0.854$	$-7.763$	$-8.232$	$-8.915$	$0.450$	$0.620$	$95.2$
	value-guided sampling	$\mathtt{RIDE}^{\text{(1)}}$	$0.340$	$0.882$	$-7.612$	$-8.157$	$-8.904$	$0.440$	$0.628$	$95.1$
		$\mathtt{RIDE}^{\text{(2)}}$	$\underline{0.334}$	$0.887$	$-7.539$	$-8.054$	$-8.806$	$0.432$	$0.621$	$93.8$
✗	random perturbation	$\mathtt{RIDE}^{\text{(1)}}$	$0.401$	$0.833$	$-7.668$	$-8.144$	$-8.794$	$\underline{0.460}$	$0.623$	$91.7$

Appendix

Table 3: Comparison of generation results with and without trajectory inversion using conDitar-a as the base model. Random perturbation corresponds to generation under

p_{\phi}(\cdot\mid t_{1}^{\ast})

, while value-guided sampling further performs search from

t_{1}^{\ast}

t=0

. Best and second-best results are shown in bold and underlined , respectively.

Explore similar work

Generating Developable 3D Molecules via Pocket-Conditioned Diffusion and Property-Aware Optimization
Jul 14, 2026Ruoxi Gao, Jiangweizhi Peng, Ziqi Chen +10Drug Design Molecular Design
Computer Science and Engineering, The Ohio State University, Columbus, OH 43210. · Industrial and System Engineering, University of Minnesota, Minneapolis, MN 55455. · Google Cloud, Google LLC, Mountain View, CA 94043. +6
SynLaD: Latent Diffusion for Generating Synthesizable Molecules Conditioned on 3D Pharmacophore Profiles
Jul 1, 2026Miruna Cretu, John Bradshaw, Patricia Suriana +6Molecules Synthesis
University of Cambridge, Cambridge, UK · Prescient Design (AI for Drug Discovery), Genentech, South San Francisco, USA · Work done during an internship at Prescient Design
PhAME: Phenotype-Aware Molecular Editing via Latent Diffusion
May 27, 2026Łukasz Janisiów, Sebastian Musiał, Bartosz Zieliński +2Molecular Optimization Drug Design
Faculty of Mathematics and Computer Science, Jagiellonian University · Doctoral School of Exact and Natural Sciences, Jagiellonian University · Jagiellonian Center for Artificial Intelligence, Jagiellonian University +2

Jul 14, 2026, cs.LGJ/K move · Enter open · S save

Generating Developable 3D Molecules via Pocket-Conditioned Diffusion and Property-Aware Optimization

Ruoxi Gao, Jiangweizhi Peng, Ziqi Chen, Frazier N. Baker+9

Computer Science and Engineering, The Ohio State University, Columbus, OH 43210. · Industrial and System Engineering, University of Minnesota, Minneapolis, MN 55455. · Google Cloud, Google LLC, Mountain View, CA 94043.+6

Drug Design Molecular Design Molecular Generation Molecules Text-Conditioned Diffusion Model Metabolic

Open paper AskPDF arXiv

Drug discovery and development is time-consuming and resource-intensive, motivating computational approaches such as diffusion models for de novo drug design. Many such models follow the structure-based drug design (SBDD) paradigm, generating molecules to fit a target binding pocket. However, existing diffusion-based SBDD methods typically couple pocket and ligand representation learning, model interactions only at the atom level, and prioritize binding affinity over other developability properties. Here, we introduce conDitar-dev, a conditional diffusion-based SBDD framework for generating ligands with strong binding affinities and favorable ADMET properties. It consists of three modules: msPRL, a pretrained multi-scale pocket representation learning module; conDitar, a pocket-conditioned diffusion model guided by msPRL representations; and paOPT, a generation-time method for optimizing ligand developability. On a newly curated benchmark of human disease targets, conDitar outperforms state-of-the-art SBDD baselines, achieving an average binding score of -8.85 kcal/mol. Across five ADMET properties, conDitar-dev improves performance by up to 73% over conDitar. To further validate the abilities of conDitar-dev to generate developable molecules, we have applied it to two validated druggable targets: programmed death-ligand 1 (PD-L1) and colony-stimulating factor 1 receptor (CSF1R) proteins. Top-ranked generatively designed molecules and their analogs have been experimentally synthesized and biologically tested. Two molecules generated directly by conDitar-dev for PD-L1 exhibited SPR-derived

K_D

values of 3.49 and 3.75

μ

M, respectively. Hit expansion based on conDitar-dev-designed molecules identified selective CSF1R inhibitors with IC

_{50}

values as low as 200 nM, while also uncovering opportunities for drug repositioning.

Abstract

Explore similar work

Generating Developable 3D Molecules via Pocket-Conditioned Diffusion and Property-Aware Optimization

SynLaD: Latent Diffusion for Generating Synthesizable Molecules Conditioned on 3D Pharmacophore Profiles

PhAME: Phenotype-Aware Molecular Editing via Latent Diffusion