Eli Lilly, among the largest pharmaceutical companies in the world, plans to use AI to identify the molecules that have the best chance of making it through the drug discovery, development and commercialization phases.
Getting a drug from one end of the often decade-long process to the other is a struggle of endurance and attrition. Only about 12% of the molecules get from phase one to commercialization, said Vipin Gopal, chief data and analytics officer at Eli Lilly, citing an industry-wide metric. The 88% of molecules that fail along the way create an opening for AI in pharmaceutical research.
"The question that you're trying to answer is, 'Can we model the properties of molecules?'" Gopal said. "And can we come up with new designs of molecules for us to test, using AI-driven methodologies?"
The pharmaceutical company aims to create AI-based models that can more rapidly detect and identify the molecules that offer the greatest potential. The goal of what Gopal refers to as "model-driven drug discovery" is twofold: Boost success rates and shrink timelines.
"It's a huge role where I think the potential of AI has not been fully explored," he said. "That is something we are working on."
Data modernization
With model-driven drug discovery on the horizon, Eli Lilly is building an organizational and technological structure for enterprise-wide AI. That task got underway about four years ago when Gopal joined the company as its first chief data and analytics officer. Among his earliest initiatives was studying Eli Lilly's data environment, which he found in need of modernization.
The question that you're trying to answer is, 'Can we model the properties of molecules? And can we come up with new designs of molecules for us to test, using AI-driven methodologies?'
Vipin GopalChief data and analytics officer, Eli Lilly
"Historically, like every other pharma company, we grew up in silos," Gopal said. "The data in the research and discovery part was managed differently compared to clinical development and manufacturing and commercial."
Eli Lilly embarked on a modernization program to create an enterprise-wide data environment, providing data quality, consistent security polices and the ability to rapidly find and access data. The program follows a cloud-first philosophy, using components such as a cloud data warehouse, to eliminate data silos and offer improved data visibility.
Eli Lilly needed such a foundation to harness AI's potential, considering, as a rule of thumb, data scientists spend 80% of their time identifying, aggregating and cleansing data, Gopal noted. The resulting data latency, as Gopal calls it, hinders AI efforts. Data latency exists in a continuum that includes analytics latency, which manual approaches exacerbate, and decision latency, which stems from the other forms.
"What we're trying to do is to reduce the data latency, and quite often that's the big one," Gopal said. Reduced data drag, coupled with an effort to speed up data analytics, translates into faster decision-making, he noted.
The bigger picture
Eli Lilly's modernization effort is part of a broader umbrella initiative, dubbed the Enterprise Data Program. Other elements include centralized analytics platforms that Gopal said help scale AI across the nearly $30 billion company that employs 36,000 people worldwide.
A natural language processing (NLP) platform, for instance, has enabled faster development with less investment than having numerous Eli Lilly analytics teams create applications independently, Gopal noted. "If all of those organizations are developing their own natural language-based technology solutions, that would be expensive and hard to scale," he added.
The Enterprise Data Program, however, isn't solely focused on platforms and cloud technologies. The initiative has also assembled a team of data scientists, which Gopal said come from a diverse backgrounds: statistics, engineering, quantitative finance, quantitative physics and computational biology.
The program is generating benefits for Eli Lilly. Gopal said his organization has built AI and machine learning models to identify the best sites for clinical trials and promote the diversity of patients participating in them. The models have significantly reduced clinical trial timelines, "getting the drugs into the hands of the patients sooner than what it would have been otherwise," he noted.
Gopal reckoned his organization is about 70% into its enterprise data and analytics trek, but he acknowledged the journey will never quite conclude.
"We always have to keep on top of making sure our data environment is modern for the algorithms and solutions that are going to be deployed," he said.
