A few decades ago, Luis Cascão-Pereira would spend countless hours in the lab refining experiments, testing enzyme after enzyme and hoping for a breakthrough. Today, IFF’s Global R&D Ventures Director inputs enzyme sequence data into an AI-driven predictive model on his laptop—from anywhere in the world—and gets answers in minutes.
With AI, we’re in a world now where we have a magic wand,” Cascão-Pereira said. “The question is: How are we using it?”
Cascão-Pereira’s journey is representative of how artificial intelligence is transforming not just research, but every aspect of IFF’s business. Powered by decades of proprietary data, AI is revolutionizing everything from enzyme engineering to global supply chains. Predictive analytics, machine learning (ML), and large language models (LLMs) are accelerating research timelines, streamlining operations and making the company nimbler.
As Cascão-Pereira was finishing up graduate school in the early 2000s, the scientific community stood on the edge of its last major scientific breakthrough: DNA sequencing. 1
The 1990-2003 Human Genome Project (HGP) was a global research consortium funded by the National Institutes of Health and the U.S. Department of Energy.2 Conducted at universities and research institutions, HGP revolutionized biomedical research by perfecting methodologies and equipment for sequencing DNA, the building blocks of all living things.3
Before the advent of cheap DNA synthesis and sequencing from the Human Genome Project, we didn’t really know where we were headed with an experiment until it was over. We were working like artisans.
Luis Cascão Pereira
Global Technology & Innovation Executive
While tried and true, this iterative approach was often unpredictable and as much of an art as a science. This made it expensive to synthesize and to sequence proteins, because scientists were often screening “blind” and only sequencing the winners or “hits” and revealing their identity.
HGP began to shift this paradigm, and within a few years of the project’s completion, biomedical research became less time-consuming and less labor-intensive, and new methodologies gave scientists more clarity at the molecular level about their research.
This enabled scientists to routinely know the identity of “everything” they were generating data for, held in site evaluation libraries (SELs). With SELs in place, scientists for the first time had enough data to start building mathematical models to explain what they were seeing and to help guide the process. These models were early examples of artificial intelligence technology, using Partial Least Squares regression (what we would call machine learning today).
Along with company-wide digitization efforts going back to 2005, this new way of working also laid the groundwork for IFF researchers to have a eureka moment of their own. Armed with new commercially available DNA sequencing tools, company scientists soon discovered they could engineer synthetic, high-performing enzymes for fabric care and dish washing.
IFF researchers started slowly, using early machine-learning tools to analyze historical laboratory data and predict what experiments would work. A major AI-driven R&D breakthrough came in the form of a liquid protease, which was the first enzyme that IFF developed with Partial Least Squares regression. Crucially, this led to cold-water proteases that would go on to become the world’s largest tonnage of a genetically engineered protein on the market for liquid laundry detergents.
“The Human Genome Project brought abundant sequencing technology to the masses,” Cascão Pereira said. “It was a watershed moment for our industry.”
Along with compressed development timelines and improved efficiency, Chief Scientist Hans de Nobel says that AI is allowing IFF to offer more personalized and effective products to its customers that enhance market position and end-user satisfaction. Specifically, IFF combines the latest data science with human-centric laboratory techniques to ensure that enzymes meet specific customer requirements, perform exceptionally well in real-world applications and can be commercially scaled.
For example, IFF’s deep reserves of protein-engineering laboratory data mean researchers can determine if an experiment is likely to succeed based on results that were previously determined in the company’s labs. Take enzyme variant stability testing. Protein LLMs can predict which variants will perform without using so much as a single test tube, a dramatic time savings when considering that up to one-third of engineering enzymes won’t survive an applied stability test.
“What sets IFF part is our reliable and fast R&D pipeline, which is enabled by our AI capabilities. Ultimately, AI means our customers get better products to market faster.”
Hans de Nobel, Chief Scientist
Take IFF’s in-development approach to activity-based costing across a complex range of activities. In formulation manufacturing, for example, AI models, in conjunction with LLMs, help break down formulation instructions into all the component processes involved: chemical processes, physical processes, and others. These models then assign costing based on variables like location of manufacturing, batch size and materials needed.
According to Alexander Manasson, Director of Data Science for NA and LATAM Operations, “This helps us identify where the big costs are and helps us pinpoint where we reduce costs to translate into reduced prices for consumers.”
While much recent media attention has focused on generative AI and LLMs, IFF also draws from a deep bench of statistical and machine learning models and optimization algorithms. These technologies allow IFF to take a data-driven approach to problem-solving, improving processes across production lines and global supply chains.
Predictive AI models for process and chemical control, for example, improve yield and throughput. As Manasson said, “This means we can cover, and deliver, to more customers faster.”
In the realm of supply chain, logistics and operations, advanced optimization algorithms help IFF increase fulfillment across a complex and large customer base. Mixed Integer Programming in the Dynamic Scheduling Optimization tool, for example, helps solve the notoriously tricky problems of optimizing production schedules based on real-time information. And linear programming in the Capacity Margin Optimization tool helps determine what product should be produced at what plant to ensure maximum satisfaction of customer demand.
“It used to take us eight hours to review 100 documents,” said Elisabeth Brackman, Senior Manager, Regulatory Business Operations. “With AI, we can review the same 100 documents in a few minutes.”
Biotechnology is a heavily regulated industry that requires IFF’s compliance team to manage filing deadlines and hundreds of thousands of documents across different IT systems and applications. Today, IFF’s deep AI capabilities and long-standing organizational commitment to digitization mean time-consuming administrative functions can be continuously identified, automated and optimized. Examples include:
To improve accuracy, consistency and compatibility across disparate systems, IFF successfully completed an automation pilot in 2023 that automatically applied metadata to dossiers at a 90% accuracy rate before human intervention. These efforts centralized metadata in a common system, reduced error-related compliance risks and enhanced findability in the company’s regulatory document repository.
To reduce manual processes, IFF successfully developed templates and tools that automated significant aspects of the dossier application process. IFF is currently identifying further aspects of the dossier submission process that AI can complete, including cataloging user requirements for redactions.
In R&D, the enzymes of the future could be used to remove heavy metals, odors and microplastics from drinking water, human skin and the environment. New design techniques could mean that biomolecules replace all traditional chemicals used in detergents so that all future home and personal care products are fully renewable and biodegradable. But don’t expect AI-powered robots to be doing this work any time soon, if ever. According to de Nobel, success in the future means that AI systems are doing the bulk of the number crunching and computation work, but the human is always in the loop, developing new modalities, assessing them and calibrating the AI’s biochemical reasoning.
“In the fully automated state-of-the-art labs of the future, humans will do human work and computers will do computer work,” de Nobel said.
These efforts, are expected to produce the first 100% biodegradable laundry and dishwasher detergent pods “in the very near future.”
Casper Vroemen, Chief R&D Officer
Casper Vroemen, IFF’s Chief R&D Officer, said the company is making big leaps in AI-driven sustainable materials science for dish and laundry applications. These efforts, he said, are expected to produce the first 100% biodegradable laundry and dishwasher detergent pods “in the very near future.”
“These efforts will add a lot to our clients’ sustainability agendas,” Vroemen said.
In summary, AI is bringing about big changes at IFF. In parallel with company-led digitization efforts, a watershed moment in the global biomedical research community more than 20 years ago created the right conditions for deploying AI across R&D, operations, and compliance and regulatory functions today. And these efforts are producing big dividends for IFF’s customers by reducing production timelines, streamlining operations and making the company more responsive to customer needs.
References
[1] https://www.genome.gov/about-genomics/educational-resources/fact-sheets/human-genome-project
[2] Ibid.
[3] Ibid.
