BiMol-Diff: A Unified Diffusion Framework for Molecular Generation and Captioning

TL;DR

BiMol-Diff introduces a token-aware diffusion framework that enhances molecular generation and captioning by preserving structurally important tokens through position-dependent noise scheduling.

Contribution

It proposes a novel token-aware noise schedule for diffusion models, improving molecular structure fidelity and captioning performance.

Findings

15.4% relative gain in Exact Match for molecule reconstruction

Achieves best BLEU and BERTScore in captioning benchmarks

Token-aware noising improves fidelity in molecular structure-language tasks

Abstract

Bridging molecular structures and natural language is essential for controllable design. Autoregressive models struggle with long-range dependencies, while standard diffusion processes apply uniform corruption across positions, which can distort structurally informative tokens. We present BiMol-Diff, a unified diffusion framework for the paired tasks of text-conditioned molecule generation and molecule captioning. Our key component is a token-aware noise schedule that assigns position-dependent corruption based on token recovery difficulty, preserving harder-to-recover substructures during the forward process. On ChEBI-20 and M3-20M, BiMol-Diff improves molecule reconstruction with a 15.4% relative gain in Exact Match and achieves strong captioning results, attaining best BLEU and BERTScore among compared baselines. These results indicate token-aware noising improves fidelity in molecular structure-language modelling.