By Thomas H. Davenport and Kimberly Alba Mc Cord
We both recently attended (and McCord spoke at) the AI Applications Summit for BioPharma, produced by Corey Lane Partners, LLC at Harvard Medical School in late October. We found the conference to be a very useful snapshot of what is happening at the intersection of AI and life sciences. We’ll describe ten of our observations below.
- Pharma firms have a big problem with drug development. It costs way too much and takes way too long—which provides motivation for the use of AI. The latest figure from the Tufts Center for the Study of Drug Development is that developing a new drug costs an average of $2.6 billion. The average time to market for a new drug is about twelve years. About 10% of drug candidates make it from Phase 1 testing to market. It is clearly worth a lot of research and spending to find out if these figures can be improved through the application of AI.
- Many companies are dabbling in the use of AI for drug development, but there are some are that are very serious about it. Among the most serious startups at the conference were Berg Health, which is gathering massive amounts of data from patients with diseases like prostate cancer (it has a partnership with the US Veterans Administration, Walter Reed Army Hospital, and Cleveland Clinic on this) in order to identify new targets and develop new drugs. Another concerted effort to use AI for drug R&D is at Insilico Medicine. The company’s CEO, Alex Zhavoronkov, spoke at the conference. Neither of us is capable of assessing the science behind Insilico’s approach, but the company has all the outward signs of hard work on AI, including many published research papers, advanced technologies like GANS (generative adversarial networks, an approach to deep learning), prominent partners, and an objective of “advanced end to end drug discovery AI.”
Startups like Berg and Insilico are moving much faster than big pharma firms, although it’s still too early to say that they will be more successful in bringing new drugs to market. According to several speakers, big pharma’s not driving the startup ecosystem—investors are. We heard both public and private complaints from startup leaders that there is lots of time-consuming tire kicking, but very little money, coming from Big Pharma. This doesn’t mean that big pharma firms aren’t using AI; we’ve heard separately, for example, that Pfizer now has over 150 AI projects underway. But few of them are at the core of drug development.
Another recurrent theme at the conference was a mixture of excitement and frustration towards AI technologies. Many companies had the tendency to jump into projects with high expectations but unclear visions of the processes and outputs of these endeavors. They seem to have learned their lessons, however. Both big pharma and startup executives agreed at the conference that model performance monitoring and better planning, including hypothesis definition and testing standards, should be priorities when setting up future AI projects. Careful project definition can also ensure that the AI implementations are not merely focused on narrow research questions or process improvements, but that a bigger picture is kept in mind in order to maintain the sustainability and promote the scalability of these projects.