Scheme 1: Groebke–Blackburn–Bienaymé (GBB) reaction.

Figure 1: Marketed drugs comprising imidazo[1,2-a]azine scaffolds.

Over the last two decades, the GBB reaction has been studied in more than 200 research papers and a number of works (including the original publication by Blackburn and coauthors ) described its parallel synthesis version . Recently, we have shown that other (pseudo-)three-component reactions are very effective for the generation of synthetically tractable ultra-large chemical space . Such virtual but readily accessible (REAL) compound libraries demonstrated excellent performance in early drug discovery programs when combined with modern computational tools such as high-throughput docking or machine learning . In this work, we aimed at the implementation of the GBB reaction for the generation of such ultra-large chemical space, including experimental evaluation of the synthesis success rate (SSR, i.e., percentage of experiments that allowed obtaining the target library member in pure form) on a large set of starting materials.

Through the article, the compound numbering system common for the works on combinatorial chemistry was used: the starting materials used for the library generation were marked as 1, 2, and 3, whereas the corresponding library members were denoted as 4.

Figure 2: Yields of library members 4 synthesized using both Sc(OTf)3 and TsOH as the catalysts.

Figure 3: Amino heterocycles 1 demonstrating poor performance in the parallel GBB reaction.

Some of these results (e.g., on the reactivity of aminopyrimidine derivatives) were in accordance with the previous literature data . Meanwhile, electronic effects of the substituents in the amino heterocycle reported in the previous works were somewhat contradictory. Whereas for the NO2 group, lowering the reactivity has been documented, other electron-withdrawing groups were reported to be generally compatible with the GBB reaction . Interference of dialkylamino or alkoxy groups at the C-6 position was also mentioned previously .

Figure 4: (Hetero)aromatic aldehydes 2 illustrating electronic and steric effects on the parallel GBB reaction.

Scheme 2: A) Parallel GBB reaction and B) examples of library members 4 obtained (relative configurations are shown).

Since validation of the GBB reaction showed that it is compatible with the main readily accessible chemical space criteria (at least 80% synthesizability) , we have aimed at the generation of the corresponding REAL space. For this purpose, 686 amino heterocycles 1, 3,927 aldehydes 2, and 107 isonitriles 3 complying with our general in-house reactivity/availability filters were subjected to virtual coupling using the limitations mentioned above. Additionally, combinations providing compounds with more than two chiral centers were not included. This resulted in 271,026,660 library members with nearly 85% expected synthetic accessibility according to the model experiments described in the previous section.

Figure 5: Physicochemical properties of the chemical space of 271 Mln. members obtained by virtual GBB reaction (MW – molecular weight; HAcc/HDon – H-bond acceptor/donor count; F(sp3) – fraction of sp3-hybrid carbon atoms; RotB – rotatable bond count); compounds complying with specific Lipinski/Veber rules (MW ≤ 500, log P ≤ 5, HDon ≤ 5, HAcc ≤ 10, RotB ≤ 10 ) as well as compounds with F(sp3) > 0.5 are highlighted in blue, the rest of the compounds are shown in yellow.

Figure 6: Distribution of maximal values among pairwise-calculated Tanimoto similarities T (MFP2 fingerprints ) of extended Bemis–Murcko scaffolds for the generated chemical space members (5.60 Mln. scaffolds) to the extended Bemis–Murcko scaffolds of A) ChEMBL compounds (v. 33); B) PubChem compounds (due to the large size of the dataset, a preliminary clusterization was performed to achieve ca. 5-fold size reduction); C) ZINC15 drug-like compounds, and D) enamine’s stock screening collection. Average T values are shown by dotted lines.

Figure 7: t-Distributed stochastic neighbor embedding (t-SNE) comparative analysis of 50,000 randomly selected molecules picked from the generated chemical space and A) ChEMBL compounds; B) PubChem compounds; C) ZINC15 compounds; and D) enamine’s stock screening collection.

Figure 8: Some biologically active representatives of the generated GBB chemical space found in the ChEMBL database.

The Groebke–Blackburn–Bienaymé (GBB) reaction, a three-component condensation of amino heterocycles, aldehydes, and isonitriles, is a powerful tool for the combinatorial synthesis of compound libraries. We have shown that the Sc(OTf)3-catalyzed version of the reaction has a wide substrate applicability and established some limitations under the parallel synthesis conditions. In particular, while the method was applicable to a wide range of aminopyridines, pyrazines, and pyridazines, it worked poorly for aminoazoles, aminopyrimidines, substrates with strong electron-withdrawing groups, and substrates bearing additional dialkylamino or alkoxy substituents. The electronic nature was a major limiting factor for other two components of the reaction, the aldehyde and isonitrile, while the steric factor was found to be not significant. The protocol was used to prepare a 790-member compound library with 85% synthesis success rate. Furthermore, a readily available (REAL) chemical space comprising 271 Mln. members was generated. It was rich in both drug-like and “beyond rule-of-five” compounds and had considerable uniqueness as compared to the available collections (as was shown by Tanimoto similarity and t-distributed stochastic neighbor embedding (t-SNE) comparative analyses). Still, 432 members of the generated chemical space were found in the ChEMBL database, and some of them had high potency against various biological targets.