None of state-of-the-art molecular embedding models surpass fingerprints
Original title: On the limitations of molecular representations
Analysis shows between 225 million and 4 billion molecules in the Enamine REAL database have non-unique representations.
Systematic investigation of molecular representations finds none of the state-of-the-art embedding models capture graph structure, 3D shape or ADMET endpoint similarity better than fingerprints or physicochemical descriptors. Classical fingerprints also revealed material limitations with non-unique representations in large databases. Findings outline directions for more robust encodings.
- No embedding model outperforms fingerprints or descriptors on key metrics
- 225M to ~4B molecules in 10.8B database have non-unique representations
- Implications for virtual screening and dataset splitting
- Evaluation often limited to narrow downstream benchmarks
From the original article
1 min read at arxiv.orgMolecular representations such as fingerprints and learned embeddings play a central role in cheminformatics and small-molecule machine learning. They serve as the primary input for a diverse array of tasks, including clustering, dataset splitting, and virtual screening, yet their evaluation is often limited to downstream model performance on narrow benchmark tasks.
By Acer Blake · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.