The promise and limitations of harnessing AI in new product development.
It’s safe to say that artificial intelligence (AI) will have a significant impact on the future of new product development. A variety of startups are already utilizing machine learning platforms to find and research new active ingredients and using this knowledge to formulate innovative new products. The major benefit of leveraging AI in this way is the speed at which new ingredients can be discovered and manufactured, compared to traditional research and development. This therefore creates a lot of enthusiasm around AI, but there is also a good amount of trepidation.
Beyond the science fiction-induced fear of sentience, more practical concerns about AI have also been proposed. Much like automation in manufacturing, will AI take away work from people? This fear is being felt particularly in content creation, and has been a major point of contention during the recent writers’ strikes in Hollywood. However, the implications of AI are far-reaching across professional fields. For example, will AI learning models replace scientists or at the very least cause companies to downsize their R&D departments?
Alina Slotnik, vice president of bioactives at Brightseed (San Francisco, CA), which utilizes a machine learning platform called Forager to discover plant bioactives and develop new products from that discovery, emphasizes that AI is a tool, and that people are still a crucial component of the process.
“I don't think of AI as a panacea or one sided. I think that people’s caution and their curiosity is really well placed and what I would say about our Brightseed approach to AI is that AI and our platform technology, Forager, is really intended to be an enabler and not a replacer of science,” Slotnik explains. “It is basically taking what scientists do and equipping them to be more focused and effective in those activities and around research as opposed to having those activities no longer be needed.”
Essentially, AI facilitates efficiency in the process of research and development. “In today's supplement world, a lot of the unique bioactives that are found are found one at a time and very iteratively,” Slotnik continues. “So, a plant source is discovered, and then that plant source is screened, and then there's a bit of a trial-and-error process to identify which health benefits that plant source has and those bioactives have the ability to affect. What Forager really does is reverse engineer that process and enables those R&D scientists to have more visibility, more transparency and more accuracy in their work, and to be able to do more compounds at a time. There will still be, I think, a critical need outside of Forager, which really identifies bioactives and delivers predictions about their value to then take that and make that real…Really, it's taking the enablement the Forager allows and using real science and real human beings to validate that in clinical trials; all of that still needs to be done by really high-quality scientists and we don't expect that Forager will replace those activities.”
Speed is one important factor, but volume is another. AI not only facilitates ingredient discovery faster, but also more ingredients at a time. At the moment, this requires companies to really narrow their energy and focus in order to not be overwhelmed.
“One of the amazing benefits but also the challenges of Forager is we humans sleep, Forager does not,” says Slotnik. “So we have a constantly expanding and multiplying source of potential bioactives for us to pursue in development. But we know that as a company we have specific resources and capabilities that need to marry up with that. We've chosen to focus our internal efforts on three specific health benefit areas around gut health, around metabolic health and weight management and around mind health, which includes sleep and cognitive benefits. We think those are three core biological health systems that underpin a lot of today's health issues and health concerns for consumers.”
Other health areas, says Slotnik, are not going unaddressed, however. Brightseed has developed partnership, and will continue to develop partnerships to complement the work it does internally. For example, recent partnerships include Blue Diamond, ADM, Kallyope, Olam Food Ingredients, and Ocean Spray.
To give you an idea of the scale of data AI can create, Andrew Franklyn-Miller, Chief Medical Officer of Nuritas, a company that leverages AI to develop new cell signaling peptides as dietary supplements from plants, explains that the company has built a library of 6 million peptides from plants over the last seven years.
“So, with our platform, we've taken multiple plant sources [of peptides], we've dried, milled and enzymatically digested them, hydrolyzed them, put them through mass spectroscopy. [Then] we've identified the exact sequences of peptides, and married them to 200,000,000 biological experiments showing up or down regulation,” says Miller. “We know that the peptides have a function and whether they work with a positive or a negative. We've also then added in layers of what would normally be preclinical work. So, are they stable? Can they survive the gut? Do they penetrate our cells? How long do they last?”
Identifying a new ingredient through Nurita’s AI platform, Magnifier, takes about 24 months, from start to finish, says Miller. Throughout this process, Nuritas has to continually refine the results. For example, they will ask Magnifier to give them a peptide that affects specific pathways in the body. The results may be over 10,000 peptides. “We can then add layers of search criteria in terms of stability, dosage size, survivability, and then we narrow the plot,” explains Miller. “Then we need to do some biological assays of those peptides that we can manufacture initially synthetically to test various functions, mainly dose and effect size on multiple pathways and check safety. [After that] we can identify them in a plant source from our library which is proprietary to Nuritas. Then we can speed up that process and select them, test their biology activity, and then [conduct] human clinicals.”
Nuritas’ goal is to develop two ingredients per year, 15 over the next 5 years, with three ingredients currently on the market. Of course, once the peptide is chosen and an ingredient is developed, the work doesn’t stop. “What we've learned in terms of scaling – we use third-party manufacturers to manufacture our products – is that refining the manufacturing process can both improve the cost of goods and also accelerate the process of accuracy and the function,” says Miller. “Furthermore, we've been able to identify new sources, via our prospecting process [which] is looking continuously at new plant sources as well as some upcycled products, We also need to do formulation work to show to our customers the versatility of the ingredients and how the clinical claims can benefit their market share."
It almost feels like AI has become real overnight, but it has been years in the making. Now that it has become feasible, the learning models will only get better and more efficient. That said, AI is still being held back by some factors.
“There is a wealth of startups in AI; all offering a large language model or a ChatGPT spin-off. The limitation is in the data” says Miller. “AI needs a training set of data in order to learn. You can create synthetic datasets which are pretend so that you can effectively bootstrap your data and get large volumes, but it's ultimately hard data that sets us apart. There are a number of publicly available data sets, and we use those to support and add to our learning, but the uniqueness of Nuritas [are the] 6,000,000 peptides that have been cataloged on 200,000,000 biological experiments. That takes time and it's progressive, so you need to have the answer to experiment one before you can proceed to experiment two. So, there's a limitation in training data because it's difficult and costly to create that data set.”
It has taken years for companies like Brightseed and Nuritas to compile and build out their libraries of bioactives and peptides to effectively harness AI and develop effective ingredients. For Nuritas, says Miller, the growth potential outweighs the development costs. Society’s problems are outpacing our ability to solve them, he explains: “Food and health is a $10 trillion industry with a $7 trillion negative each year. Malnutrition, ill health, and hospitalization costs the industry and new ingredients are very hard to find. They're very expensive and take 70 or 100 years to come into the market space. Lots of what we see today are small iterative changes of existing ingredients.”
AI changes that. The ability to quickly find new actives as well as identify multiple plant sources to find the ideal or most cost-effective source really shifts the paradigm. “Our success rate is 64% in discovering a new peptide within a four-month period, which translates to 80% clinical trial success,” says Miller.
More and more companies will certainly follow, but it may still take a while before we see widespread utilization of AI in research and development.