Science1 publisher3 min readPublished
A 960-condition robotic screen found a pseudo-seven-component branch of the Biginelli reaction
Chemists at the Institute for Basic Science mapped products across 960 condition sets of the 135-year-old Biginelli reaction and found a branch that combines seven starting molecules into one bicyclic product.
The Scientist · Science desk
What happened
- A team led by Bartosz Grzybowski at the Institute for Basic Science ran an automated robotic platform across 960 different sets of conditions for the Biginelli reaction, first reported in 1891.
- The run was set up to identify which products and pathways emerge across the whole reaction space, not to find the best conditions for a molecule the chemists already knew they wanted.
- The screen surfaced a previously unknown branch of the reaction that yields complex bicyclic structures unlike its conventional products.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- capability Scoring a screening rig for pathways instead of product yield makes it an instrument for finding molecules nobody specified in advance, on hardware a well-funded synthesis lab already owns.
- constraint A yield-scored screen has nowhere to record a product it was not asked about, so a lab running automation purely as an optimizer will not see this class of result however many wells it runs.
- decision Chemists have to decide whether to spend robot time on unscored exploration of routes already in their libraries, with one existence proof and no published hit rate to size the bet.
- precedent If a reaction with 135 years of literature behind it held an unrecorded branch, condition-space mapping of other named reactions becomes a defensible use of instrument time, and reviewers can start asking whether a familiar route was mapped or only optimized.
Conventional automated chemistry is pointed at a known product and scored on its yield [12]. A screen built that way reports essentially one number per condition set, and an unexpected product shows up in that number as a shortfall. The team at the Institute for Basic Science asked a different question of the same class of hardware: which products and pathways appear across the entire reaction space [3]. The report notes that exploring all those possibilities by hand is practically impossible [19].
The chemical AI enters after the robot. Mechanistic analysis supported by chemical AI identified the new branch as a pseudo-seven-component transformation, seven molecules of the starting components ending up in one complex product [6]. The reconstructed network then guided a redesigned synthesis that produced a family of related molecules, some approaching the structural complexity of natural products [7]. In the account published by phys.org, the platform ran conditions and the researchers redesigned the route; it does not describe a loop that chose its own next experiments [18].
The Biginelli reaction was first reported in 1891 [1]. Between that first report and this screen sit 135 years of literature [16]. The denominator here is 960 condition sets in a single reaction [2], and the phys.org account does not say how many of those sets fell into the new branch [17]. So what the work establishes is that a heavily studied reaction held a pathway that systematic condition mapping could surface [5]. The work does not put a rate on that, and the fraction of the named reactions in a process library that would repay the same treatment is still unknown.
The molecules carry the clearest measurements. Some assembled spontaneously into larger structures, with the outcome depending on concentration and temperature [8]. Others bound metal ions selectively, barium and zinc in particular, which the researchers offer as a route to selective metal sensing [9]. One compound changed its handedness preference between the solid state and solution when no metal was present [10]. Under metals it flipped: zinc ions drove like enantiomers together, barium ions favored assembly between opposite ones [11]. The report calls metal-programmable sorting of this kind extremely rare, and lists enantioselective sensing, responsive materials and molecular recognition as possible uses [13]. The uses listed are candidates; nothing here has been built into a device.
The researchers argue that hyperspace mapping could turn chemical automation from a way of speeding up experiments into a platform for discovering new chemistry [14], and that reactions studied for more than a century may still hold pathways visible only under systematic exploration [15]. The first half follows from what they did. The second is a hypothesis tested once, on one reaction, and it will take a second reaction to know whether 1891 chemistry was unusually under-mapped. In my view the cheapest place to try it is a route whose starting materials and handling a lab has already qualified, because there the marginal cost of an unscored run is mostly robot time. The paper, by Daniel Matuszczyk and colleagues, is in Nature Synthesis [4].
What to watch
- Whether another group maps the condition space of a different named reaction and reports how often a new pathway turns up.
- Whether the metal-programmable chiral sorting works inside a functioning enantioselective sensor, which the report suggests but does not build.
- Whether the full Nature Synthesis paper reports how many of the 960 condition sets landed in the new branch.