Back to Cases

    Substructure search in CHEESE Search: topological constraints for synthon-based libraries

    Substructure search: lock any motif, explore the rest of make-on-demand libraries

    A single hit from a screen is just the start. What follows it is the key part of drug design, both from the perspective of medicinal and computational chemistry: a thorough study of possible analogues and their optimization to improve hit activity and ADME properties. CHEESE makes this much more accessible than before, as it allows a thorough search of combinatorial commercial databases containing more than 18 trillion synthesizable and purchasable molecules. Substructure search then enables users to explore only the relevant chemical space: to perform scaffold hopping effectively, to fix the molecule's core and search for variants of its substituents that still retain the structural, shape-based or electrostatic features of the query, to introduce novel structural patterns, or to explore different linker variants.

    How substructure queries are specified

    Close similarity is often not the only goal a researcher pursues, as specific requirements for a molecule's constitution and substructure frequently arise. For this purpose, CHEESE Search uses topological constraints to narrow the search down to specified chemical spaces with very specific substructural features. The search runs across 8 commercial combinatorial databases and typically completes within 1–5 seconds:

    • Whole molecule: Enables a search for a specified substructure, which is treated as a subgraph. Furthermore, explicit hydrogens can be added to specify which positions must not carry a substituent, and R groups can be specified to define which positions must bear a substituent.

    • Selected substructure: Allows selecting just a part of a molecule while adding a similarity reference. This makes it possible to generate molecules that bear the selected substructure while maintaining close similarity to the reference molecule.

    • SMARTS: Makes it possible to define any SMARTS pattern that must be excluded from or included in all generated molecules.

    • Presets: Introduces a set of preset substructures that must be part of all generated molecules. A powerful feature for generating amino acid analogues or specific functional groups.

    • Linker: Allows selecting a linker within a query molecule and exploring different variants of its constitution and length, while leaving the rest of the query untouched.

    • Core: A core mark replaces one ring with any ring of the chosen size. When "Rest" is set to exactly as drawn, nothing else can change; when it is off, the rest may carry additional groups.

    Demonstration: pyrazolo[1,5-a]pyrimidines in eMolecules' space

    We used the pyrazolo[1,5-a]pyrimidine kinase chemotype as a test case. A 2025 review in RSC Advances surveys the scaffold's synthesis and its use against CK2, EGFR, B-Raf, MEK, PDE4, BCL6, DRAK1, CDK1/2 and Pim-1. The fused bicycle is rigid and planar, tolerates substitution at multiple positions, and is ATP-mimetic, which is the basis of its use as a kinase scaffold. Its medicinal chemistry is almost entirely peripheral: the core is conserved and the variation sits on the substituents, which makes it a natural scaffold-constrained query.

    The video below shows the core pattern C1=NC2N(N=CC=2)C=C1 fixed as the selected substructure, with the Zaleplon molecule kept as the similarity reference. The search was performed against eMolecules' space. The results thus explore structural analogues of Zaleplon that retain the pyrazolo[1,5-a]pyrimidine core.

    Try it yourself

    Try the search at CHEESE and let us know your feedback. We are particularly interested in which query types you use most and which presets or patterns would be most useful to you.