Molecular representations for drug design have major limitations
On the limitations of molecular representations
Machine Learning
Summary
Molecular representations are how computers understand the features of small molecules for tasks like drug discovery. The authors found that modern machine-learned ways of representing molecules don’t capture important details like their 3D shape or how they behave in the body any better than traditional methods. However, these traditional methods also fail to uniquely identify billions of molecules, which can be a problem in large chemical databases. This study shows the current representations have significant weaknesses and suggests new approaches are needed for better molecular encoding.
What this means in practice
- •For drug discovery teams: Avoid virtual screening errors caused by non-unique molecule encodings in large chemical libraries.
- •For chemical database curators: Improve molecule identification by assessing representation uniqueness to enhance database reliability.
Authors
Acer Blake, Fergus Imrie, Charlotte M. Deane
Abstract
Molecular 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. We carried out a systematic investigation of the quality of molecular representations, finding that none of the state-of-the-art molecular embedding models capture graph structure, 3D shape, or ADMET endpoint similarity better than fingerprints or physicochemical descriptors. However, classical fingerprints also revealed their own material limitations. For example, we show that depending on the representation, between 225M and $\sim$4B molecules have non-unique representations in the 10.8B Enamine REAL database, with implications for virtual screening. Our findings provide an insight into the current limitations of molecular representations, outlining future research directions for the development of more robust encodings of small molecules.