Research map / Probabilistic and scientific modeling / Molecules and proteins
Molecular Optimization Model with Patentability Constraint
Read the paperCheck your idea against 108,967 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.
Proposes MOMP, a SMILES-based multi-domain embedding translation framework that optimizes molecular properties (QED, DRD2) while keeping low patent-likeness to patented compounds.
Problem
Molecular optimization of lead compounds must improve chemical properties and similarity to the lead while producing molecules that are dissimilar from existing patented compounds (patentability constraint).
Approach
Extends embedding-embedding translation with cycle-consistent training across three molecule domains (low-property, high-property/high-PL, high-property/low-PL), and uses molecular attention over fingerprints of non-patented molecules to enforce patentability while keeping similarity to the input.
Key results
- On QED, MOMP achieves success rate 0.272 versus 0.232 for the best prior baseline (JTVAE original) in the table, with PL 0.436.
- On DRD2, MOMP achieves success rate 0.322 versus 0.147 for UGMMT with patents-loss and 0.204 for JTVAE with patents-loss.
- MOMP reports the lowest PL among models that improved the property, e.g., 0.436 on QED and 0.584 on DRD2.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add METN pre-training (ablation) | MOMP · QED optimization | Success: 0.217 -> 0.272 | Helped |
| use C-domain fingerprints instead of input fingerprints during training (ablation) | MOMP · QED optimization | Success: 0.213 -> 0.272 | Helped |
| add the Extended-EETN pathway (ablation) | MOMP · QED optimization | Success: 0.260 -> 0.272 | Helped |
| add the A ⇄ B and B ⇄ C EETN components and Extended-EETN pathway (ablation) | MOMP · QED optimization | Success: 0.225 -> 0.272 | Helped |
| add the B ⇄ C component and Extended-EETN pathway (ablation) | MOMP · QED optimization | Success: 0.088 -> 0.272 | Helped |
Surprising findings
- Higher property values tend to correlate with higher patent likeness (greater overlap with patented compounds), across single-property optimization.
Limitations
- Evaluated only on two properties (QED and DRD2).
- Patent likeness is measured against MCS from SureChEMBL rather than actual full patent sets.
Experimental setup
Datasets: Jin et al. 2019 molecule pairs (QED, DRD2) SureChEMBL
Benchmarks: QED optimization DRD2 optimization
Metrics: property score patent likeness (PL) similarity validity novelty optimization success
Baselines: UGMMT JTVAE REINVENT CORE G2G Mol-CG MIMOSA IPCA
Builds on: UGMMT IPCA patents-loss double learning
Closest papers by meaning
- Enhancing Ligand Validity and Affinity in Structure-Based Drug Design with Multi-Reward Optimization (ICML 2025 · similarity 0.86)
- Efficient Evolutionary Search Over Chemical Space with Large Language Models (ICLR 2025 · similarity 0.86)
- Molecule Design by Latent Prompt Transformer (NeurIPS 2024 · similarity 0.86)
- Beyond Drug Discovery: The Nanotechnology Molecular Optimization (NMO) Benchmark (NeurIPS 2026 · similarity 0.86)
- Uncertainty-Aware Multi-Objective Reinforcement Learning-Guided Diffusion Models for 3D De Novo Molecular Design (NeurIPS 2025 · similarity 0.86)
- Procedural Synthesis of Synthesizable Molecules (ICLR 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.