Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RTMol: Rethinking Molecule-text Alignment in a Round-trip View

Letian Chen, Runhan Shi, Gufeng Yu, Yang Yang

TL;DR

RTMol addresses the core challenge of bidirectional molecule-text alignment by introducing round-trip learning that jointly trains molecule-to-text captioning and text-to-SMILES generation within a single LLM. The framework defines a mathematically grounded round-trip metric and leverages RL with Group Relative Policy Optimization to optimize chemically faithful captions and reconstructions, bypassing noisy paired data. It demonstrates substantial improvements across multiple backbones on both round-trip and text-to-molecule tasks, and shows robustness to noisy datasets via unsupervised captioning. The approach offers a principled path toward unified molecule-text understanding with practical impact on de novo design, literature mining, and cross-modal reasoning in chemistry.

Abstract

Aligning molecular sequence representations (e.g., SMILES notations) with textual descriptions is critical for applications spanning drug discovery, materials design, and automated chemical literature analysis. Existing methodologies typically treat molecular captioning (molecule-to-text) and text-based molecular design (text-to-molecule) as separate tasks, relying on supervised fine-tuning or contrastive learning pipelines. These approaches face three key limitations: (i) conventional metrics like BLEU prioritize linguistic fluency over chemical accuracy, (ii) training datasets frequently contain chemically ambiguous narratives with incomplete specifications, and (iii) independent optimization of generation directions leads to bidirectional inconsistency. To address these issues, we propose RTMol, a bidirectional alignment framework that unifies molecular captioning and text-to-SMILES generation through self-supervised round-trip learning. The framework introduces novel round-trip evaluation metrics and enables unsupervised training for molecular captioning without requiring paired molecule-text corpora. Experiments demonstrate that RTMol enhances bidirectional alignment performance by up to 47% across various LLMs, establishing an effective paradigm for joint molecule-text understanding and generation.

RTMol: Rethinking Molecule-text Alignment in a Round-trip View

TL;DR

RTMol addresses the core challenge of bidirectional molecule-text alignment by introducing round-trip learning that jointly trains molecule-to-text captioning and text-to-SMILES generation within a single LLM. The framework defines a mathematically grounded round-trip metric and leverages RL with Group Relative Policy Optimization to optimize chemically faithful captions and reconstructions, bypassing noisy paired data. It demonstrates substantial improvements across multiple backbones on both round-trip and text-to-molecule tasks, and shows robustness to noisy datasets via unsupervised captioning. The approach offers a principled path toward unified molecule-text understanding with practical impact on de novo design, literature mining, and cross-modal reasoning in chemistry.

Abstract

Aligning molecular sequence representations (e.g., SMILES notations) with textual descriptions is critical for applications spanning drug discovery, materials design, and automated chemical literature analysis. Existing methodologies typically treat molecular captioning (molecule-to-text) and text-based molecular design (text-to-molecule) as separate tasks, relying on supervised fine-tuning or contrastive learning pipelines. These approaches face three key limitations: (i) conventional metrics like BLEU prioritize linguistic fluency over chemical accuracy, (ii) training datasets frequently contain chemically ambiguous narratives with incomplete specifications, and (iii) independent optimization of generation directions leads to bidirectional inconsistency. To address these issues, we propose RTMol, a bidirectional alignment framework that unifies molecular captioning and text-to-SMILES generation through self-supervised round-trip learning. The framework introduces novel round-trip evaluation metrics and enables unsupervised training for molecular captioning without requiring paired molecule-text corpora. Experiments demonstrate that RTMol enhances bidirectional alignment performance by up to 47% across various LLMs, establishing an effective paradigm for joint molecule-text understanding and generation.

Paper Structure

This paper contains 33 sections, 15 equations, 4 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Formulation
  4. Round-trip Metric
  5. Round-trip Optimization
  6. Model Architecture
  7. Generator: Grounding in Chemical Knowledge.
  8. Captioner: Learning via Round-trip Process.
  9. Training Strategy
  10. Choice of Optimization Algorithm.
  11. Choice of Reward Function.
  12. Coupled Training.
  13. Experiments
  14. Experimental Setup
  15. Dataset.
  16. ...and 18 more sections

Figures (4)

  • Figure 1: Limitations of current molecule-text alignment: (a) textual metrics ignore chemical fidelity; (b) captions are noisy or incomplete; and (c) separate modeling fails to enforce bidirectional consistency.
  • Figure 2: Overview of RTMol. A single LLM serves as both the Captioner and Generator for molecule-text alignment, with their training alternating cyclically in a complementary manner to reinforce each other's performance.
  • Figure 3: Cases of round-trip evaluation from the L+M-F and Mol-Instruct-F datasets. Mistakes and corrections are highlighted in red and green, respectively.
  • Figure 4: Examples of filtered high-quality molecule-text pairs from the L+M-F and Mol-Instruct-F datasets.