How chemists use AI to make the molecules that could protect tomorrow’s harvests

How AI helps shape the molecules that will help protect tomorrow's harvests

The search for new ways
to protect crops starts in the lab.

Syngenta’s chemists use Elsevier’s precision AI to choose the best way to make each new molecule, and to put the company’s own unpublished insights to work alongside a vast scientific knowledge base.

Every year, pests and diseases destroy nearly 40 percent of the world’s crops, according to the Food and Agriculture Organization of the United Nations. That is food lost before it ever reaches a table, and income lost for the farmers who grew it.

New crop protection products are part of the answer. Each one starts as a molecule in a lab, long before it reaches a field. And before anyone can test that molecule, a chemist has to work out how to make it. 

At Syngenta, which helps farmers grow resilient, healthy crops, that question comes at the very start of discovery. The answer shapes the rest of the journey: how quickly a promising molecule can be made and tested, which materials it needs, and how much solvent and waste it leaves behind.

Syngenta’s chemists tackle it with Elsevier’s AI-assisted chemistry intelligence solution. Reaxys puts more than a billion data points from chemistry journals and patents at scientists’ fingertips, including millions of reactions, so Syngenta’s chemists can build on the published work of chemists around the world.

They use it in two ways: to choose the right way to make each molecule, and to get more from Syngenta’s own unpublished experiments by combining them with that published record.

Choosing the right way to make a molecule

Most molecules can be made in several ways. Each is a different sequence of chemical reactions, which chemists call a route. Some routes are shorter. Some depend on starting materials that are hard to get. Some use solvents a team would rather avoid, or create more waste.

Once a route is chosen, lab time and materials are committed to it. The sooner a chemist can see which route is most likely to work, and at what cost, the better.

Choosing the right way to make a molecule

Most molecules can be made in several ways. Each is a different sequence of chemical reactions, which chemists call a route. Some routes are shorter. Some depend on starting materials that are hard to get. Some use solvents a team would rather avoid, or create more waste.

Once a route is chosen, lab time and materials are committed to it. The sooner a chemist can see which route is most likely to work, and at what cost, the better.

How Syngenta makes that choice

Syngenta’s Computer-Aided Synthesis Team has built an in-house platform for exactly this decision. It brings together three things: route-planning software, Syngenta’s own AI models, which predict how reactions are likely to behave, and Reaxys.

Chemists use the platform to compare complete routes side by side, whether the software suggested them, or the chemists proposed them. They weigh each one against several questions at once. Has this kind of reaction worked before? How likely is it to work here? Are the starting materials available? How much solvent and waste will it involve?

Reaxys answers many of those questions. Straight from the platform, chemists can search Reaxys for similar reactions that other chemists have already reported, check which starting materials are available, and open the papers and patents behind each result. So a suggestion from the software is never a black box. The chemist can see the published evidence behind it and judge how well it fits the problem in front of them.

The platform widens what they can see; they choose what to try.

The difference it makes

Syngenta reports that route design now takes less time. Sustainability also enters the conversation earlier, as solvent and waste can be weighed while there is still time to choose a different path, rather than discovered after the work has begun.

Getting more from what Syngenta already knows

Getting more
from what
Syngenta
already knows

Making unpublished data usable

Syngenta’s chemists have run many experiments that were never published. Together, they hold valuable knowledge about how reactions behave in Syngenta’s area of chemistry, and that knowledge is a competitive advantage

Much of that knowledge sat in electronic lab notebooks designed for people to read and search, rather than for machine learning.

So Syngenta changed how reaction data was captured at the bench.Their chemists now follow common protocols, with consistent labels and shared terms, so that every result can be found, compared and reused.

Marco Stenta, a computational chemist at Syngenta, describes the aim as building workflows where “data is captured for learning — not just recorded for compliance.”

But Syngenta’s own records could only show what its chemists had already tried. To move into the unknown, the models needed a wider view.

A broad scientific record gives the models somewhere to start

That reusable data matters most for Syngenta’s AI models: the ones inside its platform that predict how a reaction will behave and suggest which routes and conditions to try. Like any AI, they learn from examples. Syngenta’s own records are rich, but they only cover what its chemists have already tried. To make good predictions about new chemistry, the models needed far more examples.

Reaxys provided them. Syngenta trained its models in two stages. First, they learned from the curated reactions in Reaxys, picking up patterns seen across published chemistry. Then they were refined with Syngenta’s own unpublished results, so their predictions fit the company’s chemistry and the way its labs work.

And because new results are now captured in a reusable form, each experiment can feed back into the models. The knowledge they draw on grows with every reaction.

“High-quality, curated reaction data was our starting point for building reliable synthesis planning models. Think of it as the general discovery chemistry foundation — Reaxys gave us that foundation at scale. We then layered in our proprietary agrochemical knowledge, so the models speak the language of our chemistry, not just chemistry in general.”

Guillaume Berthon
Head of Digital Chemical Synthesis Crop Protection Research, Syngenta

A blueprint for the future of R&D

Syngenta’s approach reflects a wider shift. In Elsevier’s Corporate R&D of the Future 2026 report, a global survey of R&D professionals, respondents were asked what would most improve their decisions. The findings reflect the work Syngenta is doing: connect internal data with external scientific evidence, capture and reuse what the organization learns, and tailor AI to its own knowledge and workflows.

45% say important decisions are made before all the evidence is available

93% say organizations that connect internal and external data will have a competitive edge

88% say those who combine AI with trusted evidence will outperform

Across sites, each experiment can help the next

Syngenta’s platform, with Reaxys built in, is now used by the majority of chemists across its Crop Protection Research sites. Alongside faster route design, the company reports more efficient work through integrated search and stronger confidence in decisions supported by its models.

The bigger change is in how knowledge travels. More chemists start from the same broad base of published evidence, add what Syngenta has learned in its own labs, and return new results in a form others can use. What one chemist learns no longer stays with them; it becomes available to the next.

It is the kind of progress that rarely makes headlines. But it helps decide which molecules can be made, tested and taken forward, and, in time, which new products reach the farmers working to protect their crops.

“Integrating Reaxys data into our workflow for systematic access was a breakthrough moment for our strategy. It allowed us to combine trusted literature precedents with our proprietary chemistry.”

Marco Stenta
Computational Chemist, Syngenta