A machine-learning workflow developed at YOKOHAMA National University predicted both the reaction site and the reacting side in six complex diketones, despite being trained only on simpler molecules. The approach combines direct kinetic measurements from competition experiments with three-dimensional electronic descriptors calculated by quantum chemistry, offering a way to estimate chemical reactivity before a multistep synthesis is attempted.
Why product yield is not enough to predict reactivity
Chemists often use the yield of a final product as evidence that a reaction pathway is favorable. In a multistep synthesis, however, purification losses and competing side reactions can alter the isolated yield. The same molecule may also give different yields in different synthetic sequences. That makes yield a noisy proxy for the intrinsic reactivity of a particular functional group.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Organic Chemistry | $125.24 | Buy on Amazon |
| 2 |
|
Organic Chemistry (MasteringChemistry) | $319.99 | Buy on Amazon |
| 3 |
|
Organic Chemistry as a Second Language: First Semester Topics | $76.48 | Buy on Amazon |
| 4 |
|
Organic Chemistry | $65.32 | Buy on Amazon |
| 5 |
|
Organic Chemistry I For Dummies (For Dummies (Math & Science)) | $11.48 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
The YOKOHAMA team instead sought measurements that describe what happens at the reactive site itself: which of two ketone groups reacts first and how quickly the competing reactions proceed.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWhat the competition experiment measures
The researchers studied molecules containing multiple ketone groups. They mixed each substrate with varying amounts of sodium borohydride in methanol and monitored which ketone was reduced first. Because the ketone groups compete for the same reducing reagent, the outcome supplies relative reaction-rate information rather than a single end-of-synthesis yield.
#1 Best Overall
Those observations were organized into a kinetic scale, or reaction-rate map. A site that consistently reacts before its competitor is identified as the more reactive site under the specified conditions. Changing the reagent amount allows the team to characterize the competition across reaction progress instead of recording only the final product ratio.
How machine learning connects kinetics to molecular structure
The workflow adds a structural explanation to the measured rates. Quantum-chemical calculations provide three-dimensional electronic descriptors around each ketone. These descriptors encode how charge distribution and the surrounding molecular environment differ from one reactive site to another, including differences that are not captured by a two-dimensional formula alone.
Rank #2
The machine-learning model is trained to associate those descriptors with the kinetic ordering observed in the competition experiments. Its prediction target is therefore more specific than “which product has the highest yield”: it estimates both the preferred ketone and the side, or face, from which reduction occurs.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Element | What it contributes | Why it matters |
|---|---|---|
| Competition experiment | Relative information about which ketone reacts first | Measures intrinsic selectivity more directly than an isolated yield from a multistep route |
| Kinetic reaction-rate map | Orders competing sites by observed reactivity | Provides the learning target for the model |
| Quantum-chemical descriptors | Three-dimensional electronic information around each site | Connects molecular structure and environment to the measured kinetics |
| Machine learning | Maps descriptors from known compounds to predicted site and side selectivity | Enables transfer from simple training molecules to more complex targets |
What the transfer test showed
The model was trained using simple molecules, then evaluated on six more complex molecules, each containing two ketone groups. It correctly predicted both where reduction occurred and which side of the ketone reacted in all six tested molecules.
This is a narrowly defined but important result: the model transferred across a change in molecular complexity without being trained directly on those six target structures. The study does not establish universal accuracy for every ketone, reagent, solvent or reaction class; the demonstrated chemical space is the set of substrates and conditions examined in the work.
How the approach could reduce trial and error
- Measure competition directly. Run controlled reductions of a molecule with competing ketones using sodium borohydride in methanol and record which site reacts first.
- Build a kinetic reference. Repeat with varying reagent amounts to establish relative reaction-rate behavior rather than relying on one final yield.
- Calculate site descriptors. Use quantum-chemical calculations to represent the three-dimensional electronic environment around each candidate site.
- Train and apply the model. Learn the relationship between descriptors and measured selectivity, then predict the preferred site and reacting side for an untested molecule.
- Prioritize experiments. Use the prediction to choose a likely reaction site and synthetic route before committing to a longer multistep campaign.
For drug candidates, natural products and functional molecules, this could move some decisions earlier in the workflow. A chemist could identify likely chemoselectivity and facial selectivity before investing in extensive protecting-group, purification and route-optimization work. The model does not eliminate laboratory validation; it is intended to make that validation more targeted.
Rank #4
What makes the predictions transferable
Transferability comes from combining two kinds of information. Competition experiments supply a common measurement of relative reactivity, while three-dimensional electronic descriptors describe why apparently similar sites behave differently. Training on that shared relationship, rather than memorizing the names or formulas of individual compounds, gives the model a basis for applying its learned pattern to more complex structures.
Recommended Free Tools
This differs from an approach based only on product yields. Yields can incorporate every operation in a synthesis, including work-up and purification, whereas the competition measurement focuses on the ordering of reactions under a defined set of conditions. The trade-off is that new chemical classes may require additional competition experiments to establish that the descriptors and kinetic scale remain valid.
Best Value
Publication and researchers
The peer-reviewed study is published in JACS Au under DOI 10.1021/jacsau.6c01269. A YOKOHAMA National University release dated 16 October 2026 identifies Hiroaki Gotoh, associate professor in the Faculty of Engineering, as corresponding author and Daimon Sakaguchi, a doctoral student in the Gotoh laboratory, as first author. Taisei Kawasaki, Mayu Itakura and Chihiro Tada are also listed as collaborators from the university.
Gotoh described precise control of reaction rates and selectivity as fundamentally important in organic chemistry. Sakaguchi said the competition experiments connect laboratory measurements, quantum-chemical calculations and machine learning.
Limits to keep in mind
- The reported validation covers six complex diketones, not a broad survey of all organic reactions.
- Results are tied to the tested reduction chemistry and experimental conditions; changing reagent, solvent or temperature can change selectivity.
- A prediction of the preferred site and side does not by itself provide reaction yield, product stability or an optimized synthetic procedure.
- Laboratory confirmation remains necessary, especially when a target lies far outside the structures used to build the kinetic reference.
Frequently Asked Questions
Can chemical reactivity be predicted before synthesis?
This study shows that a model can predict relative ketone site and side selectivity before testing six more complex diketones, but the evidence is limited to the demonstrated substrates and reduction conditions.
Why use competition data instead of yield?
Competition data records which of two sites reacts first under controlled conditions, while a final multistep yield can also reflect side reactions, purification and other process effects.
Does the method remove the need for experiments?
No. It is designed to reduce exploratory trial and error and focus experiments; predicted reactions still require laboratory verification.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
