Dow turns to AI to accelerate chemical search

For chemists, finding just the right molecule for a particular application can be like searching for a needle in a haystack. With several million compounds to choose from, chemists often must resort to intuition when trying to solve complex problems around chemical processes.

US multinational Dow Chemical was working with a pulp and paper manufacturer to improve inefficiencies in its chemical process with a goal of producing a better, safer pulp yield. But to hit that narrow target, Dow needed a faster, more thorough method of identifying candidate molecules. So it launched a collaboration with Chemical Abstract Services (CAS), a division of the American Chemistry Society, to leverage CAS SciFinder, which, unlike generic search engines, is optimized for searching for chemical molecules from an electronic catalog of more than 200 million compounds. 

The resulting project, SmarthSearch, has earned Dow a 2023 CIO 100 Award list for IT leadership and innovation, and now enables thousands of Dow chemists to discover needed molecules in minutes that once took weeks to identify.

“This is what caused us to go down the road of finding a faster way to search molecules through a database. That was the start of this project,” says Nathan Wilmot, IT director of data client partnerships in Dow’s enterprise data and analytics group. “The innovation here is previously [chemists] relied very heavily on intuition and thumbing through catalogs to identify molecules.”

With SmartSearch for Dow, Dow chemists can very quickly filter down the “chemical, physical, and commercial availability [of molecules, as well as the] health and safety properties for a molecule or set of molecules that might fit a given application” in a matter of minutes, he says.

In its pulp pilot, for example, CAS SciFinder helped Dow identify a set of 8 to 12 safer, more sustainable molecules as possible candidates to replace the existing material the manufacturer aimed to replace.

Dow has since narrowed that pool down to two to four molecular options and expects to commercialize one of those within the next year or two, Wilmot says. “It eliminates a lot of experimentation time … and accelerates our research quite dramatically.”

A partnership based on chemistry

Chemical Abstract Service was founded in 1903 as a resources for chemists globally and is the “keeper” of a registry that contains more than 200 million substances that are indexed and curated from a variety of sources, including patents and meeting notes, according to a CAS spokeswoman.

Smart Search for Dow is a custom project that uses the CAS SciFinder proprietary search technology and a platform with a “knowledge graph” of more than 2.5 billion entities and 20-plus billion chemical relationships, according to Venki Rao, CTO of CAS.

Its content are supported by modern natural language processing (NLP) and neural network–based relevance technologies, Rao says, noting the union of its platform with new AI technologies “appended” to the existing platform provides Dow with new capabilities to power their R&D and achieve next-generation business goals.

Dow has partnered with CAS for many years but approached the company about a deeper collaboration that would unify the strengths of each to create Smart Search for Dow, says Dow’s Wilmot, who would not say whether the collaboration will extend to actual development or the use of AI in R&D but the possibilities — business and scientific — could be life altering.

There are millions of molecules that exist and finding one that has a unique use — tying the right molecules to the right application — using digital tools did not exist before and will have a profound impact on Dow’s business, Wilmot says.

“We have worked closely with them as an innovation partner specifically in this area since between 2019 and 2020 when [CAS] started to refresh its business model and have a services organization,” Wilmot says. “They’ve done a great job of collaborating with us on this and other areas.”

Next steps in computational chemistry

Dow’s 3,800 chemists continue to build their own database of chemicals and are augmenting it with CAS’s molecular modeling to further accelerate research in other areas “and make better decisions that they could not have otherwise,” Wilmot says.

SmartSearch for Dow is now being used on a number of projects across Dow’s plastics, silicone, and polyurethane businesses, as well as for many industrial solutions, the IT director notes. At any given time, Dow has 10 to 12 people committed to the collaborative solution.

“Instead of a needle in a haystack [approach], SmartSearch allows us to find the best molecule based on the data we have,” Wilmot says. “We can get this to all the researchers to find the best targets available — the most sustainable, cost-effective, high-performance, and high-value materials as quickly as they can, giving us a competitive advantage.”

All borne of what Wilmot calls “a cool collaboration between our IT organization with an external partner,” which, as it scales, will help Dow be more efficient and sustainable, he says, with potential impacts in industrial and agricultural chemistry, catalysis, drug discovery, and other areas, including identifying substitutes for replacing materials of concern, as was the case with the pulp and paper manufacturer. “That’s the critical piece for us.”