Inverse-IMPRESSION: A Graph-based Platform for Molecular Structure Elucidation from Experimental NMR Spectroscopic Properties

2026-07-10

AI Summary

This work proposes the first end-to-end graph neural network framework for automated molecular structure elucidation, designed to accurately reconstruct chemical structures of moderately complex organic molecules (≤480 Da) from real experimental one- and two-dimensional NMR spectra, including ¹H/¹³C, COSY, HSQC, and HMBC. The method leverages a reverse graph Transformer network, IMPRESSION-G2, and employs a three-stage pipeline—single-pass bond prediction, iterative structural refinement, and noise-augmented multi-round candidate generation with ensemble ranking—to achieve efficient and robust structure determination. Evaluated on simulated data, the approach attains a 77.8% accuracy for molecules with up to 30 heavy atoms. On challenging real-world experimental data, it successfully elucidates 10 out of 19 target molecules (53%), encompassing representative synthetic compounds and natural products, thereby significantly advancing NMR-driven automated structure elucidation.

Abstract

Here, we present a platform built on our inverted Graph Transformer Network, IMPRESSION-G2, which can accurately and rapidly reconstruct molecular bonding directly from experimental nuclear magnetic resonance (NMR) spectroscopic information. It comprises three interconnected stages: a one-shot model that predicts bond connectivity between atoms; a structure-correction stage that corrects the predicted structures by removing uncertain bonds and iteratively reassigning them; noise-augmented multi-shot prediction, generating an ensemble of candidate structures, which are ranked to identify the best-fit structure. By integrating a range of $^{1}$H and $^{13}$C NMR data, including two-dimensional (2D) experiments such as COSY, HSQC, and HMBC, the inverse-IMPRESSION platform correctly identifies the structures of 77.8% of molecules with up to 30 heavy atoms (H, C, N, O and F) using simulated NMR data, and 10 of 19 (53%) molecules using experimental NMR data. The experimental structures solved have molecular weights of up to 480 Da and are representative of the complex structures in synthetic and natural products that routinely challenge chemists. The inverse-IMPRESSION framework thus provides the first effective approach for automated molecular structure elucidation using graph-based machine learning on experimental data.

Problem

• molecular structure elucidation
• NMR spectroscopy
• graph-based machine learning
• automated structure determination
• experimental NMR data

Innovation

• Graph Transformer
• Molecular Structure Elucidation
• NMR Spectroscopy
• Inverse Design
• Automated Structure Prediction