Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Language-based software can help chemists evaluate reaction ideas by predicting products from reactants and reagent information. Molecular Transformer, described in a 2019 study, treats that task as translation between molecular strings: it learns patterns in recorded reactions and uses them to propose a product. Its benchmark results were strong, but they describe predictions against particular datasets—not experimental success or a guarantee for a new reaction.
What does “language-based” reaction prediction mean?
It means representing a chemical reaction as sequences of text-like symbols and training a model to map one sequence to another. Molecular Transformer takes molecular representations of reactants and, depending on the input setup, reagents, then predicts a representation of the product.
As an Amazon Associate I earn from qualifying purchases.
The analogy to translation is useful but has limits. In machine translation, a model learns patterns connecting a sentence in one language with a sentence in another. Here, the model learns statistical relationships between recorded reaction inputs and reported products. It does not reason about a reaction in precisely the same way a chemist does, and its output is a prediction based on patterns in its training data.
The method relies on SMILES, a notation that encodes molecular structures as strings. A sequence model can process those strings and learn which arrangements of symbols tend to correspond to reported reaction outcomes. The Molecular Transformer study used a transformer architecture with multihead attention, a design that can learn relationships among different parts of an input sequence.
Popular on Amazon
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
That representation is appealing because a reaction record can be presented in a form a language model can process without first turning every chemical decision into a hand-written rule. The model can examine the relationships that recur across many records, including relationships between the structures supplied as reactants, the reagent information and the product string that was reported. This flexibility is why the approach can be useful even when a chemist would describe the chemistry in mechanistic rather than linguistic terms.
Alpha Lee, the University of Cambridge researcher who led the study, described the appeal of the approach this way: “The fact that there is an underlying language in chemistry makes these methodologies very appealing and effective.” Teodoro Laino, a researcher on the project at IBM’s Zurich Research Laboratory, explained that Molecular Transformer “basically just learns the correlation between the reactant, reagent and the product, just like machine translation.” The important qualification is in the word correlation: the model learns from reaction records, and its prediction is not an experiment.
Recommended Free Tools
Can AI predict the products of chemical reactions?
Yes, systems such as Molecular Transformer can generate a likely product representation from an encoded set of reaction inputs. “Likely” is doing important work here. The software predicts the product that best fits patterns learned from its data; it does not observe a flask, measure conversion or independently verify the proposed structure.
The central task evaluated in the 2019 paper was forward reaction prediction: given reactants and reagent information, predict the product represented in the dataset. The model’s output can offer a chemist a quick proposed outcome to inspect against a reaction idea or a recorded transformation.
The paper also explored selectivity predictions and retrosynthesis. Selectivity concerns which possible product or outcome is favored when a reaction may produce alternatives. Retrosynthesis works in the opposite direction from forward prediction: it starts with a target molecule and asks what precursor transformation or disconnection could lead to it. These are related but distinct tasks, and their results should not be collapsed into one overall measure of “reaction prediction accuracy.”
Forward prediction and retrosynthesis answer different questions
For forward prediction, the model is asked what product follows from specified inputs. For retrosynthesis, it is asked what way of making a target might be plausible. A correct match to a dataset’s forward product is not the same thing as identifying every viable synthetic route, and a plausible retrosynthetic suggestion is not automatically a complete route.
A Chemistry World report from 2019 described a patent-derived retrosynthesis test in which the software found more than 43% of published disconnections correctly. That result belongs to the test described in the report. The report also discussed a separate, more focused collaborator lab-book result; the percentages from those settings should not be compared as if they came from one dataset or identical evaluation conditions.
Retrosynthesis also introduces a practical step that a headline percentage can conceal. The proposed transformations still have to be connected manually into a route. A suggestion may be useful without reproducing the exact disconnection recorded in a patent, but a chemist must decide whether successive steps are compatible, whether the proposed precursors are available and whether the route makes chemical and practical sense.
Also popular on Amazon
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
How accurate was the Molecular Transformer?
In the authors’ 2019 evaluation on the USPTO_MIT benchmark, the reported top-1 accuracy was 90.4% when reactants and reagents were separated in the input. Top-1 means the model’s first-ranked prediction matched the recorded product. In that separated-input setting, the paper reported 93.7% top-2 accuracy: the correct recorded product appeared among the model’s first two predictions.
When reactants and reagents were mixed in the model input on the same benchmark, the reported top-1 accuracy was 88.6%. The input condition matters: 90.4% and 88.6% are not interchangeable estimates of a single universal performance level. Both figures describe the authors’ evaluation on that benchmark, under the stated input arrangements.
| Evaluation | Reported result | What it means |
|---|---|---|
| USPTO_MIT, reactants and reagents separated | 90.4% top-1; 93.7% top-2 | The first prediction matched the dataset product in 90.4% of evaluated cases; one of the first two matched in 93.7%. |
| USPTO_MIT, reactants and reagents mixed | 88.6% top-1 | The first prediction matched the dataset product in 88.6% of evaluated cases with this input setup. |
| Patent-derived retrosynthesis test described by Chemistry World in 2019 | More than 43% correct published disconnections | A retrosynthesis result from a separate test; it is not directly comparable with the USPTO_MIT forward-prediction percentages. |
These numbers show that the model could reproduce recorded benchmark products frequently in the evaluated settings. They do not establish the probability that a chemist’s untested reaction will work in a particular laboratory, nor do they tell a reader how well the model performs on a different chemistry domain or a newer dataset.
Why the input condition belongs next to the percentage
Separating reactants and reagents gives the model one defined arrangement of information; mixing them gives it another. Because the reported results differ, a performance claim without the input condition loses useful context. The same discipline applies to the task itself: a forward top-1 score, a top-2 score and a retrosynthesis disconnection rate answer different questions.
Top-1 and top-2 are exact comparison measures against the product recorded in the benchmark. They are valuable for a reproducible evaluation, but they do not capture every chemically acceptable answer. A different reagent or disconnection can be sensible even when it does not exactly match the record used as the reference.
What does the uncertainty score add?
The authors evaluated a confidence score as a way to distinguish correct from incorrect predictions. They reported ROC–AUC 0.89 for this uncertainty-classification task. ROC–AUC is a discrimination measure: it indicates how well a score separates examples in which predictions were correct from examples in which they were not.
Free tools Windows power users keep installed
One-click scans. No signup required.
That result is different from top-1 accuracy. Top-1 asks whether the leading product was right; ROC–AUC assesses how well a score ranks correct cases above incorrect ones. A strong discrimination result can help identify predictions that warrant closer attention, but it does not mean that every confidence value is a calibrated probability for a new application. The evaluation does not establish calibration for an unfamiliar dataset, reaction class, or laboratory workflow.
In practice, an uncertainty signal is best used for triage. A chemist might inspect lower-confidence suggestions first, compare them with other evidence and reserve experimental planning for ideas that survive that review. A high score should not eliminate review, because the reported metric describes separation of correct and incorrect examples in the evaluated setting rather than a guarantee for each new prediction.
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
How could a chemist use a prediction responsibly?
The practical role is best understood as decision support: a proposed product can help a chemist check an idea, compare it with recorded patterns, or decide what deserves closer examination. The 2019 report quotes Lee envisioning software that could indicate whether a reaction is a “go or a no go” and suggest a way to reach a target. That is a prospective research benefit, not evidence that a prediction alone can settle whether a reaction should be run.
1. Define the question first
Decide whether the need is forward prediction, selectivity assessment or retrosynthesis. For a forward question, identify the reactants and the reagent information that will be supplied. For a retrosynthetic question, define the target and treat the output as a candidate precursor transformation rather than a finished route.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute2. Check the representation and input setup
Confirm how the molecules are encoded as SMILES and whether reactants and reagents are separated or mixed. A result under separated reactant-and-reagent input is not the same evaluation condition as one using mixed input. Recording that setup makes a later comparison with the published benchmark meaningful.
3. Use the output as a hypothesis
A predicted product is a candidate outcome to inspect, not an experimental observation. The chemist still has to judge whether the represented transformation makes sense for the intended reagents, conditions, substrate and purpose. The sources describe prediction and route suggestions; they do not establish that the software supplies experimental confirmation.
4. Compare alternatives and uncertainty
When more than one output is available, compare the leading options rather than treating the first string as uniquely correct. An uncertainty score can help prioritize review, but it should be considered alongside chemical knowledge and the limits of the evaluation. A model’s confidence cannot substitute for checking the proposed transformation.
5. Build and test the route separately
For retrosynthesis, connect proposed steps manually and assess the complete sequence. Check whether each intermediate and reagent choice is practical for the intended work. Experimental planning, execution and confirmation remain separate stages; a benchmark match does not establish yield, selectivity in the laboratory or reproducibility.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Interpret “correct” in context
Benchmark accuracy compares a prediction with the product recorded in a dataset. Chemistry World’s discussion of retrosynthesis highlights why that can be an incomplete picture: a model may suggest a different reagent or disconnection and still point to a chemically plausible transformation. Teodoro Laino noted that a prediction can include the “wrong anion in reagents” while, from a chemist’s point of view, the overall reaction is acceptable. That observation is a caution about how exact-match evaluation can differ from chemical usefulness; it is not proof that any particular alternative prediction is valid.
What are the main pitfalls?
The first pitfall is reading a benchmark percentage as an experimental success rate. A 90.4% top-1 match means the leading product string agreed with the recorded benchmark product in that evaluation. It does not mean 90.4% of future reactions will work, nor does it predict how much product will be isolated.
Also popular on Amazon
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
The second is combining unlike results. Forward prediction on USPTO_MIT, mixed-input forward prediction, and the patent-derived retrosynthesis test have different tasks, data and metrics. They should be reported with those labels intact. The collaborator lab-book discussion is another setting, not a replacement denominator for the patent test.
The third is assuming an exact mismatch is automatically a chemical failure. A model can name a different reagent or disconnection and still suggest a plausible transformation, while an exact benchmark match can remain unsuitable for a particular substrate or set of conditions. Chemical review is needed in both directions.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe fourth is treating historical results as current product claims. The 2019 study and report establish what was evaluated and discussed then. They do not establish current IBM RXN availability, present commercial terms, performance on new datasets or universal industrial impact. Anyone selecting a current service would need separate, up-to-date documentation and validation.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What do the results establish—and what remains uncertain?
The strongest supported claim is bounded: the 2019 study reported high forward-prediction accuracy on specified USPTO benchmark conditions, and it examined uncertainty scoring and selectivity as well as product prediction. The accompanying 2019 report described a retrosynthesis test and researcher views about how such tools might assist chemists.
Patent records and curated benchmark examples are not identical to every practical laboratory situation. The figures therefore should retain their dataset and input-condition labels. The cited work does not establish Molecular Transformer’s current availability, current performance on new datasets, commercial terms, or universal industrial impact. Nor does it show that a system’s proposed product has been made and experimentally verified merely because it matches a recorded benchmark entry.
For chemists, the potential benefit is a faster way to surface a data-informed reaction suggestion for review. The boundary is just as important: the model learns patterns in recorded reactions, while deciding what is chemically appropriate and establishing what happens experimentally remain separate matters.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Frequently Asked Questions
Does a 90.4% prediction accuracy mean a reaction has a 90.4% chance of working?
No. That figure measures matches to recorded products in one benchmark evaluation, not experimental success rates or yields for reactions a chemist plans to run. A benchmark match is evidence about that evaluation, not a probability for an individual laboratory experiment.
Can AI predict the products of chemical reactions?
It can generate a product representation from encoded reactants and reagent information. The result is a data-informed prediction for a defined task, not an observation or experimental confirmation.
Also popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
What is the difference between forward prediction and retrosynthesis?
Forward prediction starts with reactants and reagents and predicts the recorded product. Retrosynthesis starts with a target and proposes a precursor transformation or disconnection. A retrosynthetic suggestion still has to be evaluated and connected manually into a route.
Why are there both 90.4% and 88.6% accuracy figures?
They use different input arrangements on the USPTO_MIT benchmark. The 90.4% top-1 result used separated reactants and reagents; the 88.6% top-1 result used mixed input. They should not be treated as one universal score.
Does the 0.89 ROC–AUC mean the model is 89% certain about every answer?
No. ROC–AUC 0.89 measures how well the uncertainty score separated correct from incorrect cases in the evaluation. It is a discrimination result, not a guarantee that an individual confidence value is calibrated for a new use.
Does Molecular Transformer tell a chemist how much product a reaction will make?
The results described here concern predicting product representations, not measuring or guaranteeing reaction yield. Yield is a separate experimental outcome and is not established by the cited accuracy figures.
Can a different reagent or disconnection still be useful when it is marked wrong?
Potentially. Exact-match scoring compares with the recorded reference, while a chemist may judge an alternative transformation chemically plausible. That possibility requires expert review; it does not make every mismatched prediction valid.
Can I use the 2019 benchmark result to assess a current reaction-prediction service?
Not by itself. The 2019 findings do not establish present-day service availability or performance on newer data, so a current tool’s documentation and validation would need to be assessed separately.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →What does Molecular Transformer cost?
The cited 2019 sources do not establish current pricing, commercial terms or a purchase requirement. Cost and access would need to be confirmed from current, authoritative information rather than inferred from the benchmark paper.
The Bottom Line
Molecular Transformer showed that sequence translation can predict recorded reaction products with high accuracy in defined 2019 benchmark settings. Its outputs and uncertainty scores can help chemists prioritize ideas, but task labels, dataset conditions and experimental judgment remain essential.
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.
