AI in Drug Discovery – An AI-focused conference with precious insights from the pharma and biotech industry
Artificial intelligence (AI) has arrived in the world of drug discovery, and that’s not surprising to anyone. However, what is surprising is the speed at which it spreads transversally across the many different domains of this field – and related ones. To learn more about the impact of Machine Learning (ML) and AI in drug discovery first-hand, Eric Le Roux, Discngine CEO, and I joined the “AI in Drug Discovery” conference in London organized by the SMi group on 14th and 15th of March.
The event, chaired by the prominent expert Darren Green from GSK [1], offered an excellent overview of the different areas where AI is most extensively used in industry, including mainly R&D, but not only. Indeed, there are many fields for its application in the drug discovery pipeline and beyond. Without further ado, let’s dig into it and look at what Eric and I got from the event.
Speakers first
The 2-days agenda included impressive talks from experts with many years of experience in the drug discovery industry. It was interesting to observe that while some of them are affiliated with big pharma, a good portion of the speakers is now part of smaller industries, underlining how much room for exploration AI offers in the pharma and biotech worlds. In other words, AI provides vast opportunities especially for new startups and spinoffs to thrive and grow, increasing the variety of topics explored. This is positive because it allows fast progress and increases the whirlwind of ideas, ultimately pushing progress.
Speaking of ideas, let’s move on and see in which areas these talented people and their teams managed to harass the power of AI and ML in order to move drug discovery a step forward.
AI in the drug discovery pipeline... and beyond
In the conference program, one could already see that AI in drug discovery is not only used to predict physicochemical properties. Moreover, by listening to the different lectures, and networking with speakers, I fully realized the momentum that AI is gaining over and above the small molecules ideation and optimization cycles as well.
Even if the latter is still one of the most attractive topics – at Bayer [2] scientists have implemented an entire holistic de-novo design platform – others (some mentioned list below) are emerging as part of autonomous – or mostly autonomous – drug design platforms (examples from Benevolent AI [3] and AstraZeneca [4]). That's not only due to the ability of AI to crunch a huge amount of data regardless of the domain, but also to find cross-domain correlations and to give suggestions that we, humans, could not obtain otherwise.
Let’s now touch on the topics (some emerging, some very hot already) addressed during the conference:
Finance for biotechnology: As any other field that faces uncertainty (and drug discovery certainly is highly ranked among them), there is the need to carefully consider how to spend money and evaluate the risk/benefit at every step. During the conference, I, a computational chemist by training, was intrigued to see how AI is used to help the biotechnology industry take “smart risks” (Genentech [5]) before, during, and even after drug discovery programs.
Finding the target: One of the most interesting applications of AI in drug discovery comes indeed with the purpose of finding new targets, meaningful targets, good targets! It was fascinating listening to how companies like GSK [6], Verge Genomics [7], and Healx [8] are compiling disease pathways to then perturb them computationally, with the goal to examine how the pathology system as a whole reacts to that perturbation. This allows them to spot the hubs of the network, which, if properly modulated, could ameliorate the studied disease. Such an approach is of particular interest in the case of rare diseases where the data are generally scarce.
Finding and optimizing molecules: It is not a novelty that the pharma industry uses AI, and especially ML models, to generate ideas for new molecules and to predict their ADMET properties. However, what I think is particularly worth noting are two things:
The wide variety of ways in which a new molecule idea can be crafted from, whether other molecules, fragments (Astex Pharmaceuticals [9]), or pure de-novo design (Bayer [2]).
The level of trust that is placed in ML models for property prediction. Some of them are so reliable that a number of companies tend to not perform tests in the wet lab anymore to determine, for example, the solubility of a molecule (Merck KGaA[10]). For some cases, they now rely solely on ML-powered model predictions.
Analyzing pathology: This was indeed another “expected” topic at this conference. The omics data are vastly complicated and contain much information which an AI model can rationalize. The exciting part was to see how companies like Boehringer Ingelheim [11] and Multiomic Health [12] manage to maximize the pathology informatics with AI approaches. They can not only study the pathologies alone but also evaluate the disease treatment results. Also extremely interesting was to hear how Exscientia [13] addresses pathology investigation using image analysis – up to single cells! – in the context of precision medicine.
Clinical trials and regulatory acceptance: The COVID-19 pandemic has disrupted several drug discovery areas, and clinical trials are no exception. It was interesting to hear from Boehringer Ingelheim [14] and Servier [15] how this field has changed since then. Not only virtual clinical trials and powerful AI analysis tools to make sense of the massive amount of data collected are now available (fascinating progress in the natural language process were elucidated by Pfizer [16]), but also – and this is what I found striking – it seems that the regulatory affairs are shifting towards acceptance of the digital evidence!
The take-home message
I would wrap up with the following few words:
“AI is powerful!”
This simple – almost obvious – take-home message should not be read in the sense that AI can do what we were used to do just faster, or better, or more precisely. The power of AI lies in the fact that it gives us the possibility to address current problems from all other perspectives. We are not anymore bound to the concept of dividing the problem into small parts, solving the small parts singularly, plugging the single solutions back together, and calling it a global solution to the initial big problem. With AI-powered methods, the initial big problem can be addressed in its entirety, without the need to split it into smaller chewable chunks and lose potentially crucial pieces of information.
AI systems can exploit all the connections that would otherwise be lost, opening a world of possibilities that we are just starting to explore. The investments of big pharma and the rapidly growing number of startups and spinoffs with remarkable successes are just a testimony to this exploration. The conference has been a certification that such a trend will grow. Despite the impressive achievements already reached, my feeling is that drug discovery scientists have just started to run wild and explore the vast possibilities offered by AI and ML, and I am certainly very curious about what will come next.
I would like to thank Discngine very much for enabling me to participate in this event and in particular our CEO Eric who accompanied me during this trip. It was cool to meet our friends from Optibrium and our official Discngine partners, ChemAxon and Schrödinger.
I would also like to give kudos to all the speakers who contributed greatly to the experience, and to greet other participants with whom I had very pleasant chats during the lunch and coffee breaks!
If you like the article find it on LinkedIn and don’t hesitate to like
and comment!
Reference:
Speakers ordered according to their mention in the text:
[1]: Darren Green, GSK
[2]: Alexander Hillisch, Bayer
[3]: David Michalovich, BenevolentAI
[4]: Christian Tyrchan, AstraZeneca
[5]: James Arnold, Genentech
[6]: Patrick Schwab, GSK
[7]: Irene Choi, Verge Genomics
[8]: Neil Thompson, Healx
[9]: Carl Poelking, Astex Pharmaceuticals (UK)
[10]: Friedrich Rippmann, Merck
[11]: Varenka Rodriguez DiBlasi, Boehringer Ingelheim
[12]: Ariella Cohain, Multiomic Health
[13]: Gregory Vladimer, Exscientia
[14]: Bhupathy Alagiriswamy, Boehringer Ingelheim
[15]: Philippe Moingeon, Servier
[16]: Peter Henstock, Pfizer