Research map / Probabilistic and scientific modeling / Molecules and proteins
On the Selectivity of Generative Models in Structure-Based Drug Design
Ella Miray Rajaonson, Jungyoon Lee, William Chau, Alan Aspuru-Guzik, Benjamin Sanchez-Lengeling, Dominique Beaini, Kirill Neklyudov, Marta Skreta
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies selectivity in structure-based drug design, noting that most generative models optimize binding to a single target and ignore off-target binding due to limited training data.
Problem
Most deep generative models for structure-based drug design optimize binding to a single target and ignore selectivity against off-target proteins, which contributes to clinical attrition from side effects.
Approach
Abstract is truncated; the paper addresses selectivity in generative SBDD models, which prior work largely neglects due to limited training data with multiple protein targets. Specific method not stated in the available text.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Z-score normalization of docking scores (ablation) | docking delta-score · PARP1/PARP2 positive control | mean delta-score: -0.033 -> 0.207 | Improved |
| add Z-score normalization of docking scores (ablation) | docking delta-score · CDK9/CDK2 positive control | mean delta-score: selective -0.691 -> 0.172; non-selective -1.074 -> -0.323 | Improved |
| train for selectivity (CMD-GEN, SpecLig) vs standard SBDD models | selectivity-trained models · SelectBench-Case | top-10 delta z and selectivity rate | No clear effect |
| train for selectivity (CMD-GEN, SpecLig) vs standard SBDD models | selectivity-trained models · SelectBench-Safety-Priority | average selectivity rate and R20 | Improved |
| train for selectivity vs standard SBDD models | selectivity-trained models · SelectBench-XL | top-10 delta z and selectivity rate | No clear effect |
| use autoregressive models (Pocket2Mol, PocketFlow) instead of diffusion/flow models | autoregressive SBDD models · SelectBench-Case and Safety | selectivity metrics | Hurt |
| add inference-time guidance (CFG or FKC) to diffusion models | TargetDiff, DecompDiff, DiffSBDD · SelectBench-Case | selectivity metrics | Improved |
| add inference-time guidance (CFG or FKC) to diffusion models | diffusion SBDD models · SelectBench-Safety-Priority | R20 z and average selectivity rate | Mixed |
| add inference-time guidance | diffusion SBDD models · SelectBench-XL easy vs medium/hard | selectivity metrics | Mixed |
| use aggressive negative guidance (beta1=1, beta2=-1) (ablation) | FKC/CFG guidance · molecule validity | validity | Hurt |
| add selectivity-focused guidance | diffusion SBDD models · SelectBench-Case PARP task | potency and synthesizability | Mixed |
Limitations
- Full text unavailable; method and results not described
Closest papers by meaning
- Rethinking Specificity in SBDD: Leveraging Delta Score and Energy-Guided Diffusion (ICML 2024 · similarity 0.90)
- Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · similarity 0.90)
- CBGBench: Fill in the Blank of Protein-Molecule Complex Binding Graph (ICLR 2025 · similarity 0.89)
- Enhancing Ligand Validity and Affinity in Structure-Based Drug Design with Multi-Reward Optimization (ICML 2025 · similarity 0.88)
- FlexSBDD: Structure-Based Drug Design with Flexible Protein Modeling (NeurIPS 2024 · similarity 0.87)
- DecompOpt: Controllable and Decomposed Diffusion Models for Structure-based Molecular Optimization (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
