Research map / Probabilistic and scientific modeling / Molecules and proteins
Retrieval-based Controllable Molecule Generation
Zichao Wang, Weili Nie, Zhuoran Qiao, Chaowei Xiao, Richard Baraniuk, Anima Anandkumar
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use RetMol retrieval and fusion for QED optimization | Molecule generation with ChemFormer · QED optimization, similarity δ=0.4 | success rate: 92.8 -> 94.5 | Improved |
| Use RetMol retrieval and fusion for penalized logP optimization | Molecule generation with ChemFormer · Penalized logP optimization, similarity δ=0.6 | average penalized logP improvement: 3.73±2.85 -> 3.78±3.29 | Improved |
| Use RetMol retrieval and fusion for penalized logP optimization | Molecule generation with ChemFormer · Penalized logP optimization, similarity δ=0.4 | average penalized logP improvement: 7.71±5.65 -> 11.55±11.27 | Improved |
| Use RetMol for four-property optimization | Molecule generation with ChemFormer · GSK3β and JNK3 inhibition with QED and SA constraints | success rate: 92.3 -> 96.9 | Improved |
| Use RetMol for SARS-CoV-2 main protease optimization | Eight weak inhibitors · SARS-CoV-2 main protease optimization, similarity δ=0.6 | successful optimizations: 3/8 -> 5/8 | Improved |
| Use RetMol for SARS-CoV-2 main protease optimization | Eight weak inhibitors · SARS-CoV-2 main protease optimization, similarity δ=0.4 | average binding affinity improvement (kcal/mol): 1.67 -> 2.84 | Improved |
| Train the fusion module to predict the nearest neighbor instead of reconstructing the input (ablation) | RetMol · Unconditional molecule generation | validity: 0.834 -> 0.902 | Improved |
| Train the fusion module to predict the nearest neighbor instead of reconstructing the input (ablation) | RetMol · Unconditional molecule generation | uniqueness: 0.665 -> 0.922 | Improved |
| Use a retrieval database satisfying GSK3β and JNK3 constraints, but not QED and SA (ablation) | RetMol · GSK3β and JNK3 inhibition with QED and SA constraints | success rate: 96.9 -> 84.7 | Hurt |
| Use a retrieval database satisfying only GSK3β or JNK3 (ablation) | RetMol · GSK3β and JNK3 inhibition with QED and SA constraints | success rate: 96.9 -> 44.1 | Hurt |
| Increase the retrieval database size (ablation) | RetMol · Four-property molecule optimization | generation performance | Improved |
| Increase the number of optimization iterations (ablation) | RetMol · Four-property molecule optimization | generation performance | Improved |
Closest papers by meaning
- Geometric Representation Condition Improves Equivariant Molecule Generation (ICML 2025 · similarity 0.87)
- Learning Disentangled Equivariant Representation for Explicitly Controllable 3D Molecule Generation (AAAI 2025 · similarity 0.86)
- Molecule Design by Latent Prompt Transformer (NeurIPS 2024 · similarity 0.86)
- FlashMol: High-Quality Molecule Generation in as Few as Four Steps (NeurIPS 2026 · similarity 0.85)
- Retrieval Augmented Zero-Shot Enzyme Generation for Specified Substrate (ICML 2025 · similarity 0.85)
- Unified all-atom molecule generation with neural fields (NeurIPS 2025 · similarity 0.85)
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.
