YouTubeFacebookLinkedInTwitterinstagramrss

Engaging The Black Box: How AI Changes The Stakes For Pharma



Contributed Commentary by Richard Wendell

December 14, 2017 | In their never-ending quest to speed drug discovery, development, and commercialization, pharmaceutical and biotechnology companies are now turning to information technology’s latest darling, artificial intelligence (AI). Top firms, including GSK, Merck & Co, Johnson & Johnson, and Sanofi are diving deep. Smaller companies are also trying to become players. But how can drug developers leverage AI to reverse the decline in R&D productivity that has persisted in the face of other IT technologies, such as structure-based drug design or bioinformatics?

AI is substantively different from most IT tools. It will not be easy to retrain current IT staff to become AI-proficient data scientists. Data science is a very technical field requiring sophisticated expertise in computer science, mathematics and a new class of big data technologies. A bachelor’s degree or a few online courses will not suffice in filling these requirements. Meanwhile, the demand for AI professionals has surged in all fields. As a result, finding, and retaining qualified AI employees is a challenge.

In addition to talent challenges, embedding AI into your drug R&D process will require overhauling how your data is managed, breaking down organizational silos, and, most importantly, embracing a challenging shift in thinking.

Most companies aren’t ready to make the kind of investment required to integrate AI into operational systems. As a result, many have adopted the “toe in the water approach,” which is common when getting started with new technologies.  This tactic lets companies try out a few vendors and a few approaches without being overly invested in a single position. The idea is to lower risk while still “playing” in the field.

This approach is a good way to get started. Numerous vendors will gladly partner on targeting their respective niche within drug R&D. Many of these early players are focusing on drug discovery activities like predicting molecule – target bonding, identifying new biomarkers, and finding new drug repurposing opportunities. More recent players are starting to target opportunities in clinical. All of this innovation has led to a wider interest in AI, with Reuters recently reporting that “the world’s drug companies are turning to artificial intelligence to improve the hit-and-miss business of finding new medicines.”

Spending too long in this proof of concept phase is risky. What if such a targeted AI initiative proves successful and the key vendor is acquired? How will you transmit your petabytes of data that have accumulated on a distant computer back to your internal systems? How are you going to deal with public cloud security concerns? Companies need to ensure they can build upon any progress made out of the gates. Otherwise, after several such forays, they may find they have made no tangible progress.

So, while AI partners can quickly target a niche within the drug discovery, development, and commercialization pipeline, there are some several strategic areas that companies who are serious about using AI should begin to address immediately.

First: Get your in-house data in order. A common misconception about succeeding with AI is that one should have a better algorithm. As Peter Norvig, Google’s Research Director, famously said in 2011: "We [Google] don't have better algorithms. We just have more data." Succeeding with AI is all about having large quantities of high quality data to train the algorithm.  Unfortunately, drug developers are hampered by the fact that their data has long been stored on fragmented IT systems, missing critical metadata context, created with manual, error-prone processes, and ungoverned. This data challenge is such a long-standing issue that most pharma and biotech professionals do not know where to begin to address it.  It is addressable, and it's more important now than ever to do so, but fixing internal data issues is also a significant undertaking.

Second: Stop thinking in silos.  Due to the incredibly lengthy drug R&D process, pharma companies have broken the process down into phases (and sub-phases). An unintended consequence of this approach is that organizational silos have formed, leading to disconnecting thinking, and disconnected databases. While localized applications of AI will likely improve localized business activities, the real opportunity to reverse declining R&D productivity resides in applying AI holistically across the entire discovery, development, and commercialization cycle.  For example, many clinical phase 1 failures occur because preclinical employed an inadequate toxicity screening models.  However, one could never fix this issue without creating a single data asset that spans both preclinical and clinical, and using AI to find this pattern in the data.

Third: Be prepared for resistance to the “black box.” The most recent generation of AI technology consists of deep neural networks, which are renowned for their inability to explain why they are making a particular recommendation. For example, a neural network can easily identify pictures of cats. However, they cannot tell you how they identify cats. And this can create challenges with scientists.  Scientists are trained to understand the inner workings of every system. In fact, most scientists will say that “understanding why” is their raison d’etre. So it should come as no surprise that drug developers, who employ armies of scientists, will be challenged with adopting a technology that, in some regard, requires a mindset that runs counter to their culture.

With these pieces in place, AI can enable all kinds of decisions that may well reverse the decline in drug R&D productivity. It could turn pharma's field of silos into a seamless stream of data, enabling better decisions at every juncture, connecting those decisions across the entire system, and resulting in substantial financial returns. It's a bold vision, and, one that many are highly skeptical of, not surprisingly. Change is especially hard in an industry that is as wedded to its processes as pharma. But as someone who has seen AI transform other sectors, I believe the potential is within grasp for teams equipped with both human and artificial intelligence.

Richard Wendell is Founder and CEO of tellic LLC, a startup developing artificial intelligence (AI) tools for the pharmaceutical industry. He has over 20 years experience leading data science teams to disrupt legacy approaches to research, development, and engineering. Prior to founding tellic, he held executive and advisory roles for TE Connectivity, American Express, AT&T, and Intel. He can be reached at Richard@tellic.com.
Click here to login and leave a comment.  

0 Comments

Add Comment

Text Only 2000 character limit

Page 1 of 1