Hilbert-curve assisted structure embedding method.

Area of Science:

• Computational chemistry
• Cheminformatics
• Drug discovery

Background:

• Chemical space embedding methods are crucial for dimensional reduction, clustering, and visualization in research.
• Current methods often yield maps that are difficult to interpret and prone to rearrangement.
• These limitations hinder the insight medicinal chemists can gain into structure-property relationships.

Purpose of the Study:

• To introduce the Hilbert-Curve Assisted Space Embedding (HCASE) method for creating intuitive chemical space maps.
• To develop an embedding strategy that aligns with medicinal chemists' reasoning and logic.
• To generate robust and interpretable visualizations of chemical compound relationships.

Main Methods:

• A chemical space is defined using reference scaffolds, sorted by the Scaffold-Key algorithm.
• Scaffolds are mapped to a line, then folded into a 2D space forming a pseudo-Hilbert-Curve.
• Compound embedding is achieved by locating the nearest reference scaffold on the pseudo-Hilbert-Curve.

Main Results:

• The HCASE method generates chemical space maps organized according to medicinal chemistry principles.
• Experiments using DrugBank, CANVASS, and ChEMBL databases demonstrate the method's properties.
• The resulting embeddings are shown to be robust and intuitive.

Conclusions:

• The HCASE method provides a novel and precedential approach to chemical space embedding.
• Its use of Hilbert curves and alignment with medicinal chemist logic enhances interpretability.
• This method has the potential to improve the understanding of chemical relationships in drug discovery.