[Paper Review] De Novo Molecular Generation from Mass Spectra via Many-Body Enhanced Diffusion

Xichen Sun, Wentao Wei+3 more|arXiv (Cornell University)|Feb 2, 2026

Mass Spectrometry Techniques and Applications0 citations

TL;DR

MBGen introduces an edge-centric diffusion model with a many-body attention mechanism to generate molecular structures from MS/MS spectra, achieving state-of-the-art results on NPLIB1 and MassSpecGym benchmarks.

ABSTRACT

Molecular structure generation from mass spectrometry is fundamental for understanding cellular metabolism and discovering novel compounds. Although tandem mass spectrometry (MS/MS) enables the high-throughput acquisition of fragment fingerprints, these spectra often reflect higher-order interactions involving the concerted cleavage of multiple atoms and bonds-crucial for resolving complex isomers and non-local fragmentation mechanisms. However, most existing methods adopt atom-centric and pairwise interaction modeling, overlooking higher-order edge interactions and lacking the capacity to systematically capture essential many-body characteristics for structure generation. To overcome these limitations, we present MBGen, a Many-Body enhanced diffusion framework for de novo molecular structure Generation from mass spectra. By integrating a many-body attention mechanism and higher-order edge modeling, MBGen comprehensively leverages the rich structural information encoded in MS/MS spectra, enabling accurate de novo generation and isomer differentiation for novel molecules. Experimental results on the NPLIB1 and MassSpecGym benchmarks demonstrate that MBGen achieves superior performance, with improvements of up to 230% over state-of-the-art methods, highlighting the scientific value and practical utility of many-body modeling for mass spectrometry-based molecular generation. Further analysis and ablation studies show that our approach effectively captures higher-order interactions and exhibits enhanced sensitivity to complex isomeric and non-local fragmentation information.

Motivation & Objective

• Motivate accurate de novo molecular structure generation from tandem MS spectra.
• Address limitations of atom-centric and pairwise models by modeling many-body edge interactions.
• Develop an edge-centric diffusion framework conditioned on spectral fingerprints.
• Pretrain the spectrum encoder and the graph decoder, followed by end-to-end finetuning.
• Demonstrate improved isomer differentiation and chemical plausibility over baselines.

Proposed method

• Represent molecules as edge-centered graphs with node and edge embeddings.
• Use a spectrum encoder (MIST Formula Transformer) to produce a global fingerprint y from MS/MS data.
• Decode via a many-body enhanced diffusion process that denoises edge features conditioned on y and H.
• Incorporate a many-body attention module that updates edge embeddings via triplet interactions (i, j, k).
• Apply FiLM conditioning to inject the global spectrum information into edge representations.
• Train with a three-stage regime: spectrum encoder pretraining, decoder pretraining on fingerprint–structure pairs, and end-to-end finetuning.

Figure 1: The mass spectrum comprises fragments from complex bond cleavages, where isomers within the same peak each contribute distinct yet essential insights for accurate molecular generation.

Experimental results

Research questions

• RQ1 Can modeling higher-order edge interactions via many-body attention improve de novo molecular generation from MS/MS spectra?
• RQ2 Does edge-centric diffusion conditioned on spectral fingerprints outperform atom-centric baselines in structure generation and isomer differentiation?
• RQ3 What is the impact of pretraining the encoder/decoder and the many-body module on generation accuracy and chemical validity?
• RQ4 How well does MBGen distinguish isomeric structures with similar fragmentation patterns?

Key findings

• MBGen achieves state-of-the-art Top-1 and Top-10 accuracy on NPLIB1 and MassSpecGym benchmarks.
• On NPLIB1, MBGen Top-1 accuracy 12.20% with Top-10 22.29%, outperforming DiffMS by notable margins.
• On MassSpecGym, MBGen Top-1 accuracy 7.58% with Top-10 12.54%, outperforming DiffMS.
• MBGen yields higher MCES and Tanimoto scores, indicating better substructure and overall molecular similarity.
• Ablation studies show the many-body module and full pretraining jointly contribute to performance gains.
• Case studies illustrate MBGen’s ability to capture intra-peak isomer information and reconstruct complex molecules more accurately than baselines.

Figure 2: An overview of MBGen framework.