Research map / Vision-language and multimodal / Human and embodied modeling
RTMol: Rethinking Molecule-text Alignment in a Round-trip View
Letian Chen, Runhan Shi, Gufeng Yu, Yang Yang
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 |
|---|---|---|---|
| add RTMol round-trip RL training | ChemDFM-8B · ChEBI-20 round-trip captioning | Exact match (%): 19.0 -> 28.0 | Improved |
| add RTMol round-trip RL training | Qwen-3-8B · ChEBI-20 round-trip captioning | Validity (%): 60.0 -> 92.0 | Improved |
| add RTMol round-trip RL training | ChemT5-0.2B · ChEBI-20 round-trip captioning | Exact match (%): 12.0 -> 14.0 | No clear effect |
| add RTMol to ChemT5 for text-based molecular design | ChemT5-0.2B · ChEBI-20 text-based design | Exact match (%): 12.0 -> 13.0 | No clear effect |
| add RTMol round-trip training | ChemDFM-8B · ChEBI-20 text-based design | Exact match (%): 24.0 -> 54.0 | Improved |
| unsupervised captioning with generated descriptions instead of reference descriptions | ChemDFM + RTMol · L+M-F | Morgan similarity: 0.142 -> 0.624 | Improved |
| generated descriptions instead of reference descriptions | ChemDFM + RTMol · Mol-Instruct-F | Exact match (%): 4.3 -> 2.6; Morgan 0.196 -> 0.245 | Mixed |
| use chemical similarity reward instead of BLEU/METEOR reward (ablation) | ChemDFM + RTMol · ChEBI-20 | Morgan similarity: 0.583 -> 0.597 | Improved |
| use chemical similarity reward instead of BLEU/METEOR reward (ablation) | ChemDFM + RTMol · ChEBI-20 | BLEU: 0.735 vs 0.722 | No clear effect |
| add round-trip consistency objective (ablation) | ChemDFM · ChEBI-20 | Validity (%): 94.0 -> 98.0 | Improved |
| add round-trip consistency objective (ablation) | ChemDFM · Mol-Instruct-F | Validity (%): 82.7 -> 99.5 | Improved |
| add round-trip consistency objective (ablation) | ChemDFM · ChEBI-20 | Exact match (%): 30.0 -> 28.0 | Hurt |
Closest papers by meaning
- Atomas: Hierarchical Adaptive Alignment on Molecule-Text for Unified Molecule Understanding and Generation (ICLR 2025 · similarity 0.90)
- ReactGPT: Understanding of Chemical Reactions via In-Context Tuning (AAAI 2025 · similarity 0.87)
- KnowMol: Advancing Molecular Large Language Models with Multi-Level Chemical Knowledge (NeurIPS 2025 · similarity 0.87)
- LDMol: A Text-to-Molecule Diffusion Model with Structurally Informative Latent Space Surpasses AR Models (ICML 2025 · similarity 0.87)
- Text-Guided Molecule Generation with Diffusion Language Model (AAAI 2024 · similarity 0.86)
- TRIDENT: Tri-Modal Molecular Representation Learning with Taxonomic Annotations and Local Correspondence (NeurIPS 2025 · similarity 0.86)
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.
