Structure-based Drug Design with Equivariant Diffusion Models

TL;DR

This work addresses the challenge of structure-based drug design by introducing DiffSBDD, a $SE(3)$-equivariant 3D diffusion model that generates protein-pocket–conditioned ligands. By formulating SBDD as a 3D-conditional generation problem and enabling both de novo design and substructure inpainting, DiffSBDD achieves faithful distribution learning and flexible design constraints without retraining. The study demonstrates strong distribution alignment with ground-truth ligands, effective inpainting-based substructure redesign, and practical optimization via a lightweight evolutionary loop, including targeted on/off-target specificity improvements. The results suggest diffusion-based, geometry-aware generation can outperform autoregressive approaches and serve as a versatile, drop-in platform for multiple SBDD tasks, accelerating computational drug design workflows.

Abstract

Structure-based drug design (SBDD) aims to design small-molecule ligands that bind with high affinity and specificity to pre-determined protein targets. Generative SBDD methods leverage structural data of drugs in complex with their protein targets to propose new drug candidates. These approaches typically place one atom at a time in an autoregressive fashion using the binding pocket as well as previously added ligand atoms as context in each step. Recently a surge of diffusion generative models has entered this domain which hold promise to capture the statistical properties of natural ligands more faithfully. However, most existing methods focus exclusively on bottom-up de novo design of compounds or tackle other drug development challenges with task-specific models. The latter requires curation of suitable datasets, careful engineering of the models and retraining from scratch for each task. Here we show how a single pre-trained diffusion model can be applied to a broader range of problems, such as off-the-shelf property optimization, explicit negative design, and partial molecular design with inpainting. We formulate SBDD as a 3D-conditional generation problem and present DiffSBDD, an SE(3)-equivariant diffusion model that generates novel ligands conditioned on protein pockets. Our in silico experiments demonstrate that DiffSBDD captures the statistics of the ground truth data effectively. Furthermore, we show how additional constraints can be used to improve the generated drug candidates according to a variety of computational metrics. These results support the assumption that diffusion models represent the complex distribution of structural data more accurately than previous methods, and are able to incorporate additional design objectives and constraints changing nothing but the sampling strategy.

Figures (13)

• Figure 1: (A) Overview of the 3D diffusion setup. The diffusion process $q$ yields a noised version of the original atomic point cloud for a time step $t\leq T$. The neural network model is trained to approximate the reverse process conditioned on the target protein structure $\bm{z}^{(P)}$. Once trained, an initial noisy point cloud is sampled from a Gaussian distribution $\bm{z}_T^{(L)} \sim \mathcal{N}\left( \boldsymbol{0}, \boldsymbol{I}\right)$ and progressively denoised using the learned model. Covalent bonds are added to the resultant point cloud at the end of generation. (B) Each state is processed as a graph where edges are introduced according to distance thresholds within the ligand $d_\text{max}^{L-L}$, within the pocket $d_\text{max}^{P-P}$ and between ligand and pocket nodes $d_\text{max}^{L-P}$. (C) Replacement method for fixing molecular substructures. To complete the known part of the molecule (orange) with newly generated chemical matter (green atoms), we apply the learned denoising process to the entire molecule (orange & green), but at every step we replace the prediction for the known part (orange) with the ground-truth noised version computed with $q$. The protein context (gray) remains unchanged in every step. (D) Iterative procedure to tune molecular features. We find variations of a starting molecule by applying small amounts of noise and running an appropriate number of denoising steps. The new set of molecules is ranked by an arbitrary oracle and the procedure is repeated for the strongest candidates. (E) Antidepressant Citalopram as an example in which stereochemistry is essential for its therapeutic effect. (F) The neural network backbone is composed of MLPs that map scalar features of ligand and pockets nodes into a joint embedding space, and SE(3)-equivariant message passing layers that operate on these features and each node's coordinates. It outputs the predicted noise values for every vertex.
• Figure 2: Evaluation of distribution learning capabilities and generated examples. All targets are taken from the CrossDocked and Binding MOAD test sets. (A) Comparison of generated molecules with the reference molecule from the same pocket. We compare Tanimoto similarity of the molecular fingerprints and compute the difference $\text{Vina}_\text{gen} - \text{Vina}_\text{ref}$ between their Vina docking scores. (B) Average number of rings of different sizes per generated molecule. (C) Example molecules generated by DiffSBDD-cond for a pocket from the CrossDocked test set. We compared all generated molecules with the approximately 4.2M compounds from the Enamine Screening Collection, and selected the three closest hits with drug-likeness $\text{QED}>0.5$. Vina docking score, QED drug-likeness score and fingerprint similarity to the most similar Enamine molecules are reported in each case. (D-F) Same analyses but for target pockets from the Binding MOAD test set.
• Figure 3: Molecular inpainting design examples for scaffold hopping (A), scaffold elaboration (B), fragment merging (C), fragment growing (D) and fragment linking (E) respectively. The input to our model (the fixed atoms) is shown in blue, the outputs (designed molecules) are shown in green and the original molecule is shown in magenta for reference. PDB codes are shown for the ground truth structure. In the case of fragment merging, we compose fragments with two different crystal structures with PDB codes shown. (F) Importance of resampling for generating realistic and connected molecules. Top: Visual example; inpainted region (green) finally harmonizes with molecular context at high resamplings. Bottom: Effect of the number of resampling steps on molecular connectivity. Means and 95% confidence intervals are plotted for 3 design tasks. For this experiment we used 20 randomly selected targets from the test set.
• Figure 4: Results on molecular optimization using DiffSBDD. (A-D) Experiments on single property molecular optimization. (A) Starting molecule from PDB code 5NDU. (B) QED optimization over 8 generations. (C) SA optimization over 7 generations. (D) docking score optimization over 3 generations. We found that optimization over subsequent generations continuously optimized the docking score, but that was at expense of molecular quality. (E-G) Kinase inhibitor specificity optimization experiment. (E) Cartoon representation showing the high degree of structural similarity between our two kinases of interest (BIKE and MPSK1). (F) Trajectory plot showing the highest scoring molecule at each iteration during kinase inhibitor optimization. (G) Visual representation of the molecular graphs and bound conformations of the native and final molecules with corresponding Vina docking scores. Boxes in panels (B-D) represent the upper and lower quartile as well as the median of the data. Whiskers denote 1.5 times the interquartile range. Outliers outside this range are shown as flier points.
• Figure 5: (A) Example of a generated molecule (green) without additional resampling steps and the reference molecule (magenta) from the target PDB 5ncf. The generated molecule is not placed in the target pocket but in the protein core. (B) RMSD between reference molecules' center of mass and generated molecules' center of mass for the conditional model and inpaining model with varying numbers of resampling steps $r$. The pocket representation is $C_\alpha$ in all cases.