NewHavenBIZ

n e w h a v e n b i z . c o m | N o v e m b e r / D e c e m b e r 2 0 2 0 | n e w h a v e n B I Z 13 AI large pharmaceutical companies worldwide now report using AI, either through their own technology platforms or by partnering with smaller bioscience companies or tech firms, according to DKA. While there's currently no FDA-approved drug on the market developed solely using AI, the technology has the potential to revolutionize the drug industry over the next five to 10 years, says DKA Director Alex Creshnev. He says companies are using AI not only to find new disease treatments, but also to design clinical trials, predict their outcomes and even monitor how well their commercial products are performing with patients. "In the long run, when we collect more data about biology itself, when we know all of the biochemical processes, genetic patterns and all that, AI will be able to model the entire cellular systems and even organisms," says DKA's Creshnev. "When this becomes reality, we will be able to develop drugs without animal models, and much faster and better." BioXcel isn't the only New Haven area drug company leveraging AI. Guilford's AI erapeutics, a company founded by Connecticut serial entrepreneur Jonathan Rothberg, says it uses artificial intelligence to mine clinical, genomic and literature-based data to identify drug candidates that have already demonstrated evidence of clinical safety in patients. e computer then matches them with diseases for which they are likely to be effective. e company has already identified four drug candidates to treat diseases such as lymphoma, leukemia and amyotrophic lateral sclerosis (ALS). One of its drugs, apilimod, has shown promising anti-viral activity against COVID-19 and is now in human testing, says Dr. Murat Gunel, a Yale neurosurgeon who chairs the company's scientific advisory board. Meanwhile, BioXcel's sister company, InveniAI, also based in Guilford, has used its AI platform, AlphaMeld (which discovered BioXcel's agitation drug before the two companies split from their parent) to help companies like Alexion, Takeda, Axcella and others find new drugs. InveniAI also inked a partnership this summer with drug giant GlaxoSmithKline to help its consumer health division stay abreast of new innovations. And while most of its work has focused on drug repurposing, the company is now turning its attention to creating entirely new drug compounds using AI, says Chief Business Officer Aman Kant. e company has just started working with a Japanese biopharma to build out a pipeline of new drugs to treat gut-related disorders, he says. Connecting the dots Experts say one of the biggest benefits of AI is the potential to improve the industry's high failure rate, which can lower drug costs since the losses are oen baked into the pricing of new pharmaceuticals. Industry statistics show 90 percent of experimental drugs fail in clinical trials. "Even if you can move the needle 20 to 30 percent, that's significant in terms of bringing more drugs to the market at a faster pace — and obviously at a reduced cost because now your probability of success has gone up," says Kant. Kant says machine learning can improve the odds not only by predicting the diseases for which a drug is most likely to be effective, but also by alerting companies early on to potential safety concerns that could sink its chances for approval. For instance, he says InveniAI helped Centrexion, a Boston biotech founded by retired Pfizer CEO Jeff Kindler, to rule out any safety issues and find the best indications for three of its preclinical pain drugs before the company invested millions to advance them into clinical trials. Pharma giant Eli Lilly has since paid Centrexion $47.5 million upfront for the licensing rights to one of those drugs, a non-opioid treatment for chronic pain. While AI won't replace human expertise, it can help detect patterns and come up with hypotheses that aren't always apparent, says AI erapeutics' Gunel. He says the company's AI platform, Guardian Angel, turned up three very different diseases as likely matches for its drug apilimod: lymphoma, ALS and the Ebola virus. "at initially makes no sense," he says. "How can you have a drug that can be important for cancer, be important for neurodegeneration and be important for viral infection?" e research team later determined that the drug played a role in the cell's recycling and garbage removal system, which impacts all three of those diseases. It began human testing on the drug as a possible COVID-19 treatment this summer aer discovering that the virus used the very same pathway as a means of infecting cells. "And AI predicted this [antiviral activity] before any other experimental evidence," says Gunel. "at's the beauty of the platform. It makes those connections that are not immediately obvious. It gives you the lead, but then you have to follow the lead." AI's limitations While acknowledging its promise, some caution that AI is not a magic bullet for the drug industry. One major challenge is the lack of quality data and incomplete scientific understanding, says Derek Lowe, a drug discovery researcher based in Boston who writes a blog about the industry. "We have a lot of questions for which we don't have enough data, or for which the answers are not in the data that we have," he says. "We simply don't know enough about human biology and the biology of disease." For example, he says, no machine learning program is going to find a cure for Alzheimer's, because scientists still don't understand what causes the disease. Biotech leaders like Mehta and Gunel acknowledge the limitations, but point out that machines are constantly learning, and their predictions will only improve as scientific knowledge advances. "e thing about AI is that it keeps getting smarter and smarter as you feed it more data and as you test the data," says Gunel. n Aman Kant Murat Gunel Why biotech startups use AI for drug discovery Finding new drug targets Screening of small molecule libraries to find new drug candidates De novo drug design Drug optimization and repurposing Preclinical testing 2 8 % 4 0 % 8 % 1 7 % 7 % Source: Deloitte analysis

• Cover