The use of artificial intelligence (AI) in drug development has increased substantially over the last few years. And both large pharmaceutical companies and venture capital are starting to take notice.

In June, Takeda Pharmaceutical Co. Ltd. formed a multiyear agreement with Numerate Inc., of San Bruno, Calif., to develop drugs for oncology, gastroenterology and central nervous system disorders using Numerate's AI platform that not only finds potential targets, but can also model absorption, distribution, metabolism and excretion as well as toxicity of the drug candidates.

Two months later, London-based Astrazeneca plc signed up Berg LLC, of Boston, to use Berg's Interrogative Biology AI platform to help develop drugs to treat neurological disorders such as Parkinson's disease. Berg will use Astrazeneca's curated library of central nervous system optimized fragments that can be screened and validated with Berg's AI platform that compares the genome, proteome, lipidome and metabolome of healthy and diseased cells.

And last week, Recursion Pharmaceuticals Inc., of Salt Lake City, announced a research collaboration agreement with Osaka, Japan-based Takeda to search for drug candidates for rare diseases using Recursion's AI architecture. The pact includes an up-front payment and success-based milestone payments to Recursion of more than $90 million if the collaboration produces multiple drug approvals as well as single-digit royalties on net sales of the drug.

The deal follows an agreement between Recursion and Sanofi SA's Genzyme last year to screen Genzyme's clinical-stage small molecules for new purposes with Genzyme having an option to develop products for any new indications identified. Financial terms of that agreement weren't disclosed. Chris Gibson, co-founder and CEO of Recursion told BioWorld Insight that the company has other partnerships with large companies as well that it isn't allowed to disclose.

In addition to funding from partners, earlier this month, Recursion announced an infusion of $60 million through a series B financing led by the Silicon Valley-based venture firm Data Collective. (See BioWorld, Oct. 4, 2017.)

Recursion uses typical phenotypic screening process with cells in wells of multiwall plates exposed to a variety of compounds. The power comes in the amount of data Recursion can create and process, measuring about 1,000 features in each well with fluorescent micrograph images.

The company is generating 2 million new images and 20 TB of data each week at a pace that's accelerating as it goes along. "We've generated a majority of the data in the last six months," Gibson said, while pointing out that he could make that same comment at any point during the last two years.

Recursion started with rare diseases but has expanded into other disease areas, looking for the hundreds of changes in the features it measures that are different compared to normal cells. "We're looking for the drug that makes most of those changes go back to normal," Gibson explained. "We don't have to have a hypothesis of what pathway is going to fix it."

In addition to partners' libraries, Recursion has bought libraries to increase its screening capacity. Since it's staining for the same 1,000 features, the company can limit the number of compounds it continues to screen by throwing out duplicate compounds that create the same changes. Gibson noted that when it buys a new library it might only end up keeping 10 percent of the compounds after the initial screening.

Recursion has an ambitious goal of creating 100 new treatments that are in the clinic or approved by 2025. It's well on its way with about 30 programs in its pipeline, although most of those are still in the in vitro stage of testing.

Being focused on discovery, Recursion plans on partnering all of its compounds for at least the next 18 to 24 months. Some of the pipeline candidates will get licensed in one-off deals that can generate up-front cash, but Recursion is also exploring the possibility of spinning drugs into new companies supported by venture capital, allowing it to retain upside through partial ownership in the startup.

"There are advantages of having a portfolio of both," Gibson said. "It's about building a portfolio so we get short- and long-term revenue."

Recursion's ultimate goal is to use the vast amounts of data to build a map of human cellular biology that would allow it to predict how new compounds will respond. Gibson said the company can already do predictions that are "moderately useful" and expects that in three to five years it'll be able to better predict what could be hits.

Venture capital buys in

Last week, venture capital firm Menlo Ventures brought on Greg Yap as a partner to help dole out 15 percent of its $450 million Menlo XIV fund to companies working at the intersection of computers and life sciences. Menlo has already invested in Recursion and Cofactor Genomics Inc., which is using RNA sequencing for drug discovery as well as tumor characterization.

Menlo Ventures has primarily invested in consumer and enterprise technologies, although it has dabbled in life sciences in the past, including incubating Gilead Sciences Inc. Yap, who was most recently entrepreneur-in-residence at Illumina Ventures, where he focused on new genomics investments and was co-founder and CEO of a stealth mode digital health company called Pyrames, said he thinks the multidisciplinary approach on the VC side makes sense because he expects the teams Menlo invests in will also be multidisciplinary.

While Yap is clearly pro-AI, he said he worries that the term is becoming a "buzz word" that is deviating from its true meaning of machine learning. Some people are using it to identify when computers are used to process large amounts of data without any additional higher-level analysis, Yap explained, repeating a joke he recently heard that AI now stands for "algorithm inside."

Yap said he sees an opportunity in AI drug development to build additional dimensions on the collected data, which will allow the companies to build platforms that can be "extremely prolific" compared to single-asset companies.

Of course, with the ability to create a large number of drug candidates comes with the challenge of deciding the right balance of preclinical and clinical data to maximize returns when handing them off to a partner. "That challenge has not changed in the industry," Yap said.