TeraShapeSearch: MedChemica’s Contribution to Ultra Fast Shape Searching

Over the last thirty years using the shape and colour similarity of molecules has been an important tool in lead discovery chemistry. The overarching concept that molecules with similar shape and electronic profile will bind similarly to proteins is a foundational concept in medicinal chemistry and could be considered a more physical view of the concept of bioisosterism.

A recent advance in screening has been the availability of truly massive virtual compound libraries from make-on-demand services such as Enamine REAL with sizes in the trillions. Recent work suggests that the larger the database searched for a given query, the more likely you are to find a molecule similar to your query.¹

Therefore, the ability to search for molecules by their shape and “colour” within virtual databases becomes a useful tool in virtual screening. To search the largest virtual databases, the approach of pre-searching the synthons that are used to make a library can be used.² Recently this has been made available in the RDKit for both substructure and fingerprint searching.³ The concept of introducing shape searching to ultra large libraries has also been introduced.⁴

As part of the Openbind initiative, a key step is after running a fragment screen to then to grow, merge or join fragments to generate a scaffold with better potency. This extended molecule should ideally recapitulate the binding characteristics of the parent fragments. Of the three fragment-to-scaffold strategies, fragment joining remains the most technically demanding, especially where fragments have a significant separation. The challenge is where there is a large separation between fragments, the new linker must create a conformation that maintains the fragments in the same relative geometry while using chemistry that is applicable to the fragments.

We considered this an excellent place to apply shape-based searching of large libraries as we could imagine that a query of two (or more) fragments in a synthon encoded library could work by finding matching shape representatives of each fragment in each synthon set and only where both fragment matches were present then enumerating the molecules and testing if the overall shape of the fully ’joined’ molecule recapitulated the fragment queries. Such hits can then be bought and tested, avoiding the complexities of searching to see if chemistries can be envisaged that could connect the original fragments in a whole molecule.

An additional benefit is that “negative shapes” can be encoded. Volumes of space round the fragments where if any hit molecule overlaps, it is penalised. This enables us to convert a fragment shape search into a virtual screening tool where the “negative shapes” represent areas of a protein cavity where there would be a clash with the binding site.

In his recent RDKit Blogpost, David Cosgrove describes how this has been reduced to practice, the technical issues overcome and how it can be used as open source code in the RDkit for all to use.

This was brought together as a joint effort by ourselves at MedChemica, Dave at CozChemIx and Fergus Imrie at the University of Oxford and funded by the OpenBind Consortium.

(1) Liu, F.; Mailhot, O.; Glenn, I. S.; Vigneron, S. F.; Bassim, V.; Xu, X.; Fonseca-Valencia, K.; Smith, M. S.; Radchenko, D. S.; Fraser, J. S.; Moroz, Y. S.; Irwin, J. J.; Shoichet, B. K. The Impact of Library Size and Scale of Testing on Virtual Screening. Nat Chem Biol 2025, 21 (7), 1039–1045. https://doi.org/10.1038/s41589-024-01797-w.
(2) Liphardt, T.; Sander, T. Fast Substructure Search in Combinatorial Library Spaces. J. Chem. Inf. Model. 2023, 63 (16), 5133–5141. https://doi.org/10.1021/acs.jcim.3c00290.
(3) Cosgrove, D. Introducing Synthon Searching – RDKit blog. Introducing Synthon Searching – RDKit blog. https://greglandrum.github.io/rdkit-blog/posts/2024-12-03-introducing-synthon-search.html (accessed 2026-04-14).
(4) Cheng, C.; Beroza, P. Shape-Aware Synthon Search (SASS) for Virtual Screening of Synthon-Based Chemical Spaces. J. Chem. Inf. Model. 2024, 64 (4), 1251–1260. https://doi.org/10.1021/acs.jcim.3c01865.

Discover our other projects here.