Research map / Probabilistic and scientific modeling / Molecules and proteins
LIMO: Latent Inceptionism for Targeted Molecule Generation
Peter Eckmann, Kunyang Sun, Bo Zhao, Mudong Feng, Michael Gilson, Rose Yu
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 |
|---|---|---|---|
| Predict properties from the decoded molecular representation instead of directly from latent z (ablation) | LIMO · QED and penalized logP maximization | optimization score: penalized logP 6.52 -> 10.5; QED 0.910 -> 0.947 | Improved |
| Predict properties from decoded molecules rather than directly from latent z (ablation) | LIMO · 1,000-molecule unseen test set | r^2: 0.04 -> 0.38 | Improved |
| Replace the fully connected VAE encoder and decoder with 8-layer, 512 hidden dimension LSTMs (ablation) | LIMO · QED maximization | maximum QED: 0.3 | Hurt |
| Add a self-attention layer after the LSTM encoder (ablation) | LIMO with LSTM encoder · Molecule generation performance | performance | No clear effect |
| Use LIMO rather than the strongest reported baseline | Penalized logP maximization · Penalized logP maximization (top 1) | penalized logP: 45.0 vs 10.5 | Hurt |
| Use LIMO rather than GCPN | LogP targeting (-2.5 < logP < -2.0) · LogP targeting | success (%): 85.5 vs 10.4 | Hurt |
| Use LIMO rather than GCPN for logP targeting | LogP targeting (-2.5 < logP < -2.0) · LogP targeting | diversity: 0.392 -> 0.914 | Improved |
| Apply LIMO for similarity-constrained p-logP maximization at δ=0.0 and δ=0.2 | Similarity-constrained p-logP optimization · δ=0.0 and δ=0.2 | p-logP improvement: δ=0.0: 7.0 ± 1.4 -> 10.1 ± 2.3; δ=0.2: 5.1 ± 1.8 -> 5.8 ± 2.6 | Improved |
| Apply LIMO for similarity-constrained p-logP maximization at higher similarity thresholds | Similarity-constrained p-logP optimization · δ=0.4 and δ=0.6 | p-logP improvement: δ=0.4: LIMO 3.6 ± 2.3; δ=0.6: LIMO 1.8 ± 2.0 | No clear effect |
| Use LIMO rather than GCPN for docking-based affinity optimization | Single-objective binding affinity maximization · ESR1 | K_D (nM): 6.4 -> 0.72 | Improved |
| Optimize binding affinity together with QED and SA instead of affinity alone | LIMO · ESR1 ligand generation | drug-likeness and synthesizability | Improved |
| Use LIMO rather than GCPN for multi-objective ligand generation | Multi-objective binding affinity optimization · ESR1 and ACAA1 | K_D (AutoDock-GPU, nM): ESR1: 810 -> 4.6; ACAA1: 8500 -> 28 | Improved |
Later papers that compared with this method
1 later accepted paper reports LIMO as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Hierarchical Structure-Property Alignment for Data-Efficient Molecular Generation and Editing (AAAI 2026 · better on 1 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MF-LAL: Drug Compound Generation Using Multi-Fidelity Latent Space Active Learning (ICML 2025 · similarity 0.87)
- Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration (NeurIPS 2025 · similarity 0.86)
- Molecule Design by Latent Prompt Transformer (NeurIPS 2024 · similarity 0.85)
- Rethinking Specificity in SBDD: Leveraging Delta Score and Energy-Guided Diffusion (ICML 2024 · similarity 0.85)
- TurboHopp: Accelerated Molecule Scaffold Hopping with Consistency Models (NeurIPS 2024 · similarity 0.85)
- DeltaDock: A Unified Framework for Accurate, Efficient, and Physically Reliable Molecular Docking (NeurIPS 2024 · 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.
