Annotix
Overview
Annotix addresses one of the core challenges in untargeted metabolomics: annotating unknown molecules from mass spectrometry data (LC-MS/MS). The approach uses molecular networking — exploiting spectral similarity between compounds to propagate known annotations to structurally related unknowns.
Partners
| Institution | Role |
|---|---|
| Observatoire Océanologique de Banyuls | Marine chemistry, LC-MS/MS data, scientific lead |
| SCAI | ML pipeline, molecular networking |
My Role
Lead the ML and data science work.
Responsibilities: pipeline design, graph ML method selection, cheminformatics component supervision, coordination with chemists on domain requirements.
Technical Approach
- Spectral similarity — pairwise cosine and modified cosine similarity between MS/MS spectra
- Molecular network — graph where nodes are spectra and edges reflect spectral similarity above a threshold
- Label propagation — propagating known annotations to unknown compounds via graph-based semi-supervised learning
- Graph ML — GNN-based (Graph Convolutional Networks) prediction of molecular classes from spectral embeddings
Stack: Python · Matchms · NetworkX · RDKit
Scientific Context
Molecular networking (popularised by GNPS) groups related compounds visually, but annotation still relies heavily on manual curation. Annotix aims to automate this at scale, with a focus on marine natural products — a chemically rich and underexplored domain.
Status
🟡 In progress