Hypothesis-and-Refinement Learning of Organic Structures from Multimodal Spectroscopic Data

英文原文与中文译文逐段对照 · English original with Chinese translation

Chengchun Liu#, Zhiyuan Yan#, Li Yuan*, Hao Li, Boxuan Zhao, Yonghong Tian, Bartosz A. Grzybowski*, and Fanyang Mo*

刘丞军#、晏志远#、袁粒*、李昊、赵博轩、田永鸿、Bartosz A. Grzybowski* 和莫凡洋*

Cite This: Angew. Chem. Int. Ed. 2026, e9212238. DOI: 10.1002/anie.9212238. Preprint: https://arxiv.org/abs/2607.19816 (arXiv: 2607.19816, submitted 22 Jul 2026)

引用本文: Angew. Chem. Int. Ed. 2026, e9212238. DOI: 10.1002/anie.9212238. Preprint: https://arxiv.org/abs/2607.19816

ABSTRACT: Determining molecular structures from spectroscopic data remains fundamentally challenging because the inverse problem is intrinsically underdetermined: individual spectra are sparse, low-dimensional, and encode only partial structural evidence relative to the vast space of possible molecules. We address this challenge by formulating automated structure elucidation as a scalable hypothesis-refinement paradigm that tightly integrates spectral evidence with large-scale molecular priors. To supply structure-resolving NMR signals for multimodal learning, we construct QM9SPIN, a DFT-derived dataset comprising diverse 1D and 2D spectra, including J-coupling, DEPT experiments, and explicit spin-spin interactions. On this foundation, we introduce SpectroMol, a spectrum-to-structure model that proposes chemically valid molecular hypotheses conditioned on multimodal spectral inputs. Complementarily, we develop MS-Mol2Mol, a high-resolution mass-constrained molecular generator that integrates molecular formula, exact mass, and degree of unsaturation within a conditional generative prior trained on 400 million molecules, ensuring global compositional consistency and chemically realistic refinement. The integrated system achieves 93.8% top-1 accuracy on the simulated benchmark, adapts effectively from simulated to experimental spectra with limited experimental fine-tuning, and further improves experimental predictions through mass-guided refinement, establishing a scalable route toward automated, data-driven organic structure elucidation.

摘要:从光谱数据确定分子结构在根本上仍具挑战性，因为这一逆问题本质上是不适定的：单个谱图稀疏、低维，且相对于海量可能的分子空间仅编码了部分结构证据。我们将自动化结构解析构建为一个可扩展的"假设—精修"范式，将谱图证据与大规模分子先验紧密整合，以应对这一挑战。为给多模态学习提供可解析结构的核磁共振（NMR）信号，我们构建了 QM9SPIN——一个基于 DFT 的数据集，包含多样的 1D 与 2D 谱图，涵盖 J 耦合、DEPT 实验以及显式的自旋—自旋相互作用。在此基础上，我们提出 SpectroMol，一种谱图到结构的模型，可在多模态谱图输入条件下提出化学合理的分子假设。与之互补，我们开发了 MS-Mol2Mol，一种高分辨率、受质量约束的分子生成器，将分子式、精确质量和不饱和度整合到一个基于 4 亿个分子训练的conditional生成先验中，从而保证全局组成一致性并实现化学上逼真的精修。该集成系统在模拟基准上取得了 93.8% 的 top-1 准确率，能够以有限的实验微调有效地从模拟谱图迁移到实验谱图，并通过质量引导的精修进一步提升实验预测，从而为自动化、数据驱动的有机结构解析提供了一条可扩展的路径。

注：本文已在 Angewandte Chemie International Edition 上线，全文双语正文待补充。期刊正式版本；预印本见 arXiv:2607.19816。