Structure-informed deep generation enables de novo metabolite annotation in untargeted metabolomics

2026-04-20 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, long

Summary

MetGenX, a novel structure-informed encoder-decoder neural network, addresses the challenge of de novo metabolite annotation in untargeted metabolomics by generating metabolite structures directly from MS2 spectra. Published on April 20, 2026, MetGenX reformulates the spectrum-to-structure task as a structure-to-structure generation problem, leading to improved accuracy and chemical space coverage. In independent tests, it achieved a top-1 accuracy of 55.9% on 1388 NIST MS2 spectra and 68.5% on 1681 spectra from real biological samples, outperforming existing in silico tools. Its design ensures robust performance across both positive and negative ionization modes without requiring retraining. A multi-step annotation workflow using MetGenX successfully identified two previously uncharacterized metabolites in mouse liver untargeted metabolomics data, which were absent from major human metabolome databases.

Key takeaway

For metabolomics researchers struggling with identifying unknown metabolites, MetGenX offers a significant advancement by providing a highly accurate, structure-informed deep generation tool. You should consider integrating MetGenX into your untargeted metabolomics workflows to enhance de novo annotation capabilities and accelerate the discovery of novel chemical entities, especially for compounds not present in existing databases.

Key insights

MetGenX uses a structure-informed deep neural network to generate metabolite structures directly from MS2 spectra, improving annotation accuracy.

Principles

• Reformulate spectrum-to-structure as structure-to-structure generation.
• Structure-informed design ensures robust performance across ionization modes.

Method

MetGenX employs an encoder-decoder neural network to generate metabolite structures from MS2 spectra, leveraging a structure-informed approach to enhance accuracy and chemical space coverage.

In practice

• Apply MetGenX for de novo metabolite annotation.
• Utilize MetGenX for discovering uncharacterized chemical entities.

Topics

• MetGenX
• Metabolite Annotation
• Untargeted Metabolomics
• MS2 Spectra
• Encoder-Decoder Neural Network

Code references

• ZhuMetLab/MetGenX

Best for: AI Scientist, Research Scientist

Related on AIssential

• Integration of alternative fragmentation techniques into standard LC-MS workflows using a single deep learning model enhances proteome coverage — A unified deep learning model and integrated MS platform enhance proteome coverage by leveraging diverse fragmentation techniques.
• Generative approaches to kinetic parameter inference in metabolic networks via latent space exploration — Latent space exploration enables repurposing trained generative networks for targeted metabolic dynamics without retraining.
• Molecular deep learning at the edge of chemical space — Unfamiliarity, a reconstruction-based metric, quantifies molecular distribution shifts and predicts model reliability for novel drug discovery.
• Fine-Tuning Biological Foundation Models with LoRA Using NVIDIA BioNeMo Recipes — LoRA enables efficient fine-tuning of large biological foundation models on single GPUs by adapting a minimal parameter subset.
• 🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub — Scaling protein language models with vast, diverse data reveals emergent biological capabilities and enables programmable design.
• Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching — Atom ordering in neural representations can encode chemical knowledge, improving retrosynthesis model efficiency and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.