By Allison Proffitt, Editorial Director, AI Trends
At last week’s DECODE: AI for Pharmaceuticals forum, pharma leaders discussed the cultural challenges of AI in pharma and what steps their institutions are taking to better incorporate AI in the enterprise.
Editor’s Note: Learn more about the DECODE event, read an interview with Dominie Roberts, Cambridge Innovation Institute Senior Event Director, and Emma Huang, Senior Director of Data Sciences External Innovation at Johnson & Johnson Innovation, and a member of the DECODE advisory board.
First, Puneet Batra, director of machine learning at the Broad Institute, argued that pharma research—biology specifically—has a crucial role to play in driving AI and computing research. His work is part of the new Eric and Wendy Schmidt Center at the Broad, which has as its mission to position biology to drive the next era of computing.
Puneet Batra, Director, Machine Learning, Broad Institute
Batra identified two great revolutions of the 21st Century: the explosion in data technologies (machine learning, cloud, etc.) as well as the blossoming of biological technologies (sequencing, single-cell genomes, medical imaging, etc.). These two revolutions are converging, but the goal is not simply to apply machine learning to biological questions.
Machine learning, thus far, has been driven by image recognition and predictive accuracy, Batra pointed out. Machine learning needs to move from predictive accuracy to causal modeling, addressing “why” questions instead of only “what” questions. Biology and its unique biological questions should be a key driver to advances in computing.
Biological questions come with some specific constraints that will shape new machine learning and computing strategies. Data aren’t available at unlimited scales, data reduction runs risks of losing biological complexity, and models applied in the clinic demand a heightened level of scrutiny. But Batra thinks these are the very drivers that should be shaping computing in the future. The goal, he said, is “to make the central questions biology needs to address, this causal aspect, this mechanistic aspect, to make those key needs drivers of additional advances in computing.”
What Data, Which Problems
The question-focused approach was a theme throughout the event. Start with a question in mind, several pharma leaders argued in a panel, instead of starting with the data at hand. People tend to focus first on data or algorithms, said Paul Bleicher, founder of PhaseForward, most recently at Optum Labs, and now principal at Evident Health Strategies.
This approach misses the more fundamental question: What problem are you seeking to solve and how—if solved—will that create value or quality for the business and the patients. Only then, Bleicher said, you begin to ask: “What data would you need? Which of the datasets that we have access to can be used? When will that data potentially create bias? Where will it create issues? Once you have that all together, figure out what algorithms and the way you’ll put it all together.”
This problem-first approach enables you to think clearly about how much—and what kind—of data you actually need and which tools you’ll use to process it. Be careful of spending all of your available time, money, and resources getting datasets so beautifully cleaned that there is no bandwidth left for using and acting on the data.
Jacob Janey, Scientific Director, Bristol-Myers Squibb
Jacob Janey, scientific director, chemical and synthetic development at Bristol-Myers Squibb, argued for a minimum viable model approach to both the data needed and the algorithm chosen. Get “good enough” data, which will depend heavily on the question you are seeking to answer or the problem you hope to solve, he said. And then choose an analysis option that is sufficient for its purpose. “People tend to jump to deep learning or neural nets when sometimes it could be a simple regression or a simple random forest, which has its own benefits,” he said.
Reimagining the AI Org Chart
Reza Olfati-Saber, PhD, Global Head AI & Deep Analytics, Digital & Data Science R&D, at Sanofi outlined the organizational structure that will undergird a true AI-enabled pharma company. He proposed a pyramid architecture with computing (cloud, infrastructure) as its wide base, advancing through applications (data storage, app development, security), data (data governance and security), analytics (data analytics and visualization), machine learning, and finally AI policy (quality and ethics).
Olfati-Saber argues that pharma’s data and AI enterprise should be led by a top digital expert and a top AI expert working together. It is “practically impossible” to expect a Chief Data Officer to know the entire pyramid well enough to facilitate a digital transformation, he said. The tag-team approach is essential. “Anything else wouldn’t do the job,” he said.
Sessions from DECODE: AI for Pharmaceuticals, are now available on demand.