Chemical reactions are often written as simple equations: starting materials go in, and a product comes out. In reality, the samesubstrates can follow many different pathways depending on their concentrations, temperature, catalysts, and other conditions.
Exploring all of these possibilities by hand is practically impossible. Now, advances in laboratory automation and chemical artificial intelligence are allowing researchers to systematically map this vast “reaction hyperspace”—and uncover chemistry that may have remained hidden even in reactions studied for more than a century.
To address this question, scientists led by Prof. Bartosz A. GRZYBOWSKI, Director of the Center for Algorithmic and Robotized Synthesis within the Institute for Basic Science (IBS) used an automated robotic platform to explore 960 different sets of conditions for the Biginelli reaction, a classic multicomponent reaction first reported in 1891.
Instead of searching for the best conditions to produce a known molecule, the researchers set out to identify the different products and reaction pathways that could emerge across the entire reaction space. Their search uncovered a previously unknown branch of the Biginelli reaction that produces complex bicyclic structures unlike its conventional products.
Mechanistic analysis supported by chemical AI revealed that the unexpected pathway corresponds to a pseudo-seven-component transformation, in which seven molecules of the starting components ultimately contribute to the formation of one complex product. Guided by this newly reconstructed reaction network, the researchers then redesigned the synthesis and produced a family of related molecules, including structures approaching the architectural complexity of some natural products.
The newly discovered molecules were notable not only for their structural complexity, but also for their unusual supramolecular behavior.
Some of the compounds spontaneously assembled into larger structures in ways that depended on concentration and temperature. Others selectively bound metal ions, particularly barium and zinc, suggesting potential applications in selective metal sensing.
One compound showed an especially unusual form of chiral self-sorting. Molecules can exist as mirror-image forms known as enantiomers, and these forms do not always interact in the same way. In the absence of metal ions, the compound showed different preferences for assembling with molecules of the same or opposite handedness depending on whether it was examined in the solid state or in solution.
The behavior could also be controlled by the identity of the metal ion. In the presence of zinc ions, molecules with the same chirality preferentially associated with one another, while barium ions favored assembly between opposite enantiomers. Metal-programmable chiral sorting of this kind is extremely rare and could be useful in areas such as enantioselective sensing, responsive materials, and molecular recognition.
The broader significance of the work lies in how the reaction was discovered. Conventional automated chemistry is often used to optimize the yield of a predetermined product. Here, the robotic platform was instead used to map the reaction network itself, allowing unexpected products and pathways to emerge from regions of chemical space that would normally remain unexplored.
The researchers argue that this “hyperspace” approach could transform chemical automation from a tool for speeding up experiments into a platform for discovering entirely new chemistry. Even reactions that have been studied for more than a century may still contain hidden pathways that become visible only when their conditions are explored systematically.
The study demonstrates that combining robotic experimentation, large-scale reaction mapping, and chemical AI can reveal not only new reaction mechanisms, but also structurally complex molecules with unexpected functional properties.
This study was published in Nature Synthesis.