Precision Medicine in Pharma: Sanofi’s Frank Nestle

Frank Nestle, Sanofi’s world head of analysis and chief scientific officer, was impressed to enter the well being sciences subject after studying an Albert Camus novel and realizing his calling was to assist others. In his present position, Frank oversees the pharmaceutical firm’s transition from main care to specialty care, which incorporates creating medicines for immunology, oncology, and uncommon illnesses. On this episode of the Me, Myself, and AI podcast, Frank explains how synthetic intelligence allows Sanofi to work towards drug discovery in additional agile methods.

Learn extra about our present and observe together with the sequence at https://sloanreview.mit.edu/aipodcast.

Subscribe to Me, Myself, and AI on Apple Podcasts, Spotify, or Google Podcasts.

Give your feedback on this two-question survey.

Transcript

Sam Ransbotham: Synthetic intelligence has the potential to scale drug discovery like by no means earlier than. Learn how one pharma firm makes use of AI on at the moment’s episode.

Frank Nestle: I’m Frank Nestle from Sanofi, and also you’re listening to Me, Myself, and AI.

Sam Ransbotham: Welcome to Me, Myself, and AI, a podcast on synthetic intelligence in enterprise. Every episode, we introduce you to somebody innovating with AI. I’m Sam Ransbotham, professor of analytics at Boston Faculty. I’m additionally the AI and enterprise technique visitor editor at MIT Sloan Administration Overview.

Shervin Khodabandeh: And I’m Shervin Khodabandeh, senior associate with BCG, and I colead BCG’s AI follow in North America. Collectively, MIT SMR and BCG have been researching and publishing on AI for six years, interviewing lots of of practitioners and surveying hundreds of corporations on what it takes to construct and to deploy and scale AI capabilities and actually remodel the way in which organizations function.

Sam Ransbotham: Immediately, Shervin and I are speaking with Frank Nestle, world head of analysis and chief scientific officer at Sanofi. Frank, thanks for speaking with us. Welcome.

Shervin Khodabandeh: It’s very nice to fulfill you, Dr. Nestle.

Frank Nestle: Hello, Shervin. Good assembly you, too.

Sam Ransbotham: Frank, perhaps begin along with your present position at Sanofi. What are you doing?

Frank Nestle: I’m the worldwide head of analysis and CSO [chief scientific officer] at Sanofi. Sanofi is mostly a firm making medicines out there the world over, throughout 100 nations, with roughly 100,000 staff, offering medicines and vaccines actually to hundreds of thousands of individuals.

We’re at present going by means of a really thrilling transformation; we’ve been initially in main care and now are transitioning into specialty care, offering medicines in immunology, oncology, uncommon illnesses — together with uncommon hematology illnesses — neurology, in addition to vaccines. And my position as world head of analysis is to essentially uncover after which translate the following era of breakthrough medicines to sufferers, who’re actually ready, I can let you know that.

Shervin Khodabandeh: Inform us a bit about your background, Frank — how you bought began in your profession and the way you ended up right here.

Frank Nestle: I’m a clinician and a scientist. I educated initially in dermatology and scientific immunology, allergology. However I used to be at all times pushed by the search to make a distinction to sufferers, and I did that by making an attempt to grasp illness mechanisms after which translating these mechanistic insights into therapeutics — hopefully, and ideally, exactly tailor-made to the wants of a person affected person or a affected person inhabitants. Now it’s referred to as precision drugs, however that’s what at all times drove me.

And by way of the science, I attempted to handle to get to these illness mechanisms. One is the science of the immune system. And I’m pleased to say, with perhaps all of your listeners now, that through the pandemic, all of us grew to become immunologists.

Shervin Khodabandeh: Due to Google.

Frank Nestle: So there’s a number of know-how on the market about what’s the distinction between an antibody, a T-cell, and a B-cell.

Sam Ransbotham: I’m unsure there’s a number of know-how. There’s a number of consultants.

Frank Nestle: So, a lot of consultants, and that makes really doing the science I do much more enjoyable as a result of you’ll be able to completely dominate any dinner dialog. The opposite matter I may simply speak about for fairly some time is dermatology, which is at all times a pleasant matter throughout dinner rounds.

Nevertheless it’s extremely fascinating to consider the immune system. It’s primarily a group of cellular cells circulating by means of our physique, and so they transfer forwards and backwards between tissues the place most cancers occurs — autoimmune illness occurs — and the blood. They usually’re actually accessible through a easy blood draw, and it’s fairly thrilling to grasp how this linked system — my sturdy perception is all biology is linked — is enjoying out in most cancers, for instance, to guard us from most cancers, and we’ve got big successes in checkpoint immunotherapy of most cancers, but additionally if it will get uncontrolled, if it overreacts by way of autoimmunity. And the elemental hardwiring for the immune system is basically what we apply and exploit in vaccines. Our immune system has advanced to battle pathogens.

Shervin Khodabandeh: Inform us how the brand new world versus the outdated world of immunology is being reworked by expertise, presumably by higher analytics, higher AI.

Frank Nestle: There’s really a terrific story [about] how the good convergence of life sciences with knowledge sciences and engineering has performed out to extend our data in regards to the immune system, and all of it converges on the subject of single-cell immunology. We are able to now assess and analyze the immune system of a affected person with just some thousand cells, and we do that by making use of single-cell immunology applied sciences, the place we will research 2,000 genes per cell. So take into consideration the magnitude of gene modifications we will research if we take 100,000 cells, and throughout these 100,000 cells, we research 2,000 genes per cell. And this has been attainable due to progress in engineering, microfluidics. It has been attainable due to AI. For instance, at Sanofi, at our institute, we’re writing code, we’re writing an AI algorithm, to then primarily analyze these knowledge and to rediscover cell fates, and we annotate, then, if a cell is a B-cell or a T-cell or some new immune cell we by no means ever heard about.

The thrilling reality can be that we cannot solely get cells from the blood, from circulation; we will go into, for instance, the joints of a affected person with arthritis, and even the cerebrospinal fluid of a affected person with a number of sclerosis. So rapidly, we will go into the compartments the place illness performs out after which actually put this illness beneath … you possibly can name it a molecular microscope.

And if you concentrate on the truth that the cell is the frequent denominator of a physiological system, of the immune system, then you possibly can name it primarily the atoms of the immune system. We are able to research the immune system actually at atomic-level decision.

Shervin Khodabandeh: And the usage of AI right here is each to grasp the mechanism in addition to to perhaps provide you with new medication?

Frank Nestle: Precisely. So the primary half is, I’m at all times saying, first you must discover and create a map, a panorama, of physiology and of pathology, and that is what we’re at present doing. After which, upon getting all that perception, you create the speculation behind it — how the mechanism performs out — and then you definitely make medication.

First, we need to perceive how a illness mechanism performs out, and that’s precisely the place we use these AI algorithms. Think about you may have 100,000 cells, and so they have 2,000 genes up or down, and we don’t know if it’s a T-cell, a B-cell, or some new immune cell. So when you run a studying algorithm on these knowledge units, that algorithm will get higher and higher to truly inform us, so that you primarily reorganize, or rediscover, the mobile annotation of the immune system.

Shervin Khodabandeh: It looks as if in some functions of AI that we — Sam and I — have been speaking about, it’s an present enterprise course of that will get improved a bit, or optimized a bit. And, Frank, you’re speaking about functions that wouldn’t even exist with out AI, I’d assume, as a result of the dimensions that you simply’re speaking about, with hundreds of cells and hundreds of genes, it seems to me these would have been unsolvable issues with out the usage of AI.

Frank Nestle: I are likely to agree. At all times, the query is, “What do you imply by AI?” There’s an entire spectrum, from AGI, synthetic basic intelligence — and we’re not there but — to numerous functions of machine studying, the place you … After I began off — a few years in the past — to make use of machine studying approaches, these have been simply easy clustering algorithms or these have been, for instance, random forest-type machine studying approaches. However now we will clearly do far more by way of making use of AI.

The invention of medicines is then an entire totally different order of problem. And that is the place, primarily, you get into this query of molecular design. Medicines come in numerous dimensions and shapes. In the event that they’re under 500 daltons, then they’re referred to as small molecules, or above, they’re giant molecules. Small molecules are your typical drugs you’re taking, and huge molecules may very well be biologics: antibodies you inject. And as we all know from COVID, you’ll be able to inject antibodies to guard you from COVID, however you may as well take small-molecule drugs to mainly block, for instance, the viral replication of the COVID virus.

So how do you get to these molecules? You first have to grasp your goal. So, for instance, if you concentrate on a goal in a number of sclerosis or in systemic lupus, you may have a sure goal; it’s a protein. You must perceive the construction of that protein, and that is referred to as structural enablement. We are able to now use improbable applied sciences, corresponding to cryo-electron microscopy or X-ray decision of these targets, and we will uncover their construction.

However what we will additionally do is, we will run just about actually billions of small molecules in opposition to these targets after which uncover small molecules hitting these targets in a manner that it has a practical influence. And these are usually referred to as then allosteric inhibitors, so an inhibitor which primarily binds to a goal, to a protein, and does so by way of an output that’s practical. So it could actually then, for instance, block a sure goal, block a sure protein.

And that is then step one to a medication as a result of we’ve got, rapidly, a goal. We’ve got a illness the place it performs a task. We’ve got a device compound — for instance, a small molecule. After which we will take this beginning chemistry to finally then get to an actual drugs.

Sam Ransbotham: You’re speaking about making an attempt all these items. Are you making an attempt them in actual time, within the bodily world, or are you making an attempt them just about?

Frank Nestle: It’s a mix of the true and the [virtual] world. I’m at all times speaking about AI because the chemist across the desk. For instance, if you wish to provide you with a brand new construction of a small molecule, a medication, we used to do crowdsourcing. We’ve got greater than 300 chemists in our group, so you possibly can ship that crowdsourcing request on the market, and you’d have 300 chemists’ brains engaged find the right molecule.

However when you’ve got AI at play and use generative algorithms, then you’ll be able to actually undergo not hundreds of thousands however lots of of hundreds of thousands of potential buildings after which optimize them. It’s simply an order of magnitude bigger of what we used to do. For instance, we did high-throughput screening simply with a number of hundred thousand molecules. Now we will do it with lots of of hundreds of thousands of molecules, and we will do that just about.

However then that digital display will get you simply to a success — we name it the primary iteration of a possible drugs — after which this must be optimized. And it’s really this cross discuss between the machine and the human, which is going on on a regular basis.

Shervin Khodabandeh: And the human on this case is a chemist?

Frank Nestle: The human can be a chemist, yeah, precisely.

Shervin Khodabandeh: So chemists nonetheless have a job right here?

Frank Nestle: Completely. Completely.

Sam Ransbotham: Yeah, I used to be going to ask what that crowd of chemists thinks, that all of a sudden 300 chemists aren’t requested for his or her enter. What do they give thought to this?

Frank Nestle: Precisely. So they really benefit from the problem. Simply merely what we name the ligandable house, or the chemical house, we’re exploring is rising always. And thru the rise of that chemical house, we will provide you with utterly new molecules. There’s nothing extra attention-grabbing for a chemist to be confronted with than a brand new molecule they haven’t even considered after which put that into movement.

Nevertheless it’s an extended street from this unique hit to finally a scientific candidate. It takes usually 4 to 5 years. After which you must clinically translate this; this takes one other eight years. So it’s an extended journey from this unique hit, however that is the place, primarily, a molecule is born. A medication is then born later within the clinic, after we do proof-of-principle research in sufferers.

Sam Ransbotham: It appears a bit of bit tough although. For those who’ve bought all of a sudden many, many extra hits, doesn’t that create a big effect in your course of and your workflow downstream from that?

Frank Nestle: You would possibly suppose so, however really the reverse performs out.

Sam Ransbotham: Oh, actually?

Frank Nestle: So our expertise tells us that these fashions, they predict buildings, and so they assist us to scale back that big house to a a lot smaller house. So I’ll provide you with an thought: We usually needed to synthesize about shut to five,000 molecules with the everyday drug discovery paradigm we have been making use of a number of years in the past. For those who use the help of a predictive mannequin, like an AI mannequin — and there are a number of steps within the worth chains to get to a molecule to finally apply fashions — you’ll be able to cut back this to 500.

And that is the massive promise now — and it’s actually vital to grasp that — is that predictive energy of an AI algorithm being fed and educated with a number of knowledge units, refining the variety of molecules we have to conceptually provide you with after which check in our assay system. So really, what it does is it reduces the funding we’d like for synthesis — and it’s costly to synthesize compounds — however then additionally to check these compounds. And finally what this could result in is that we cut back the timelines. And when you perceive the arithmetic or the economics of drug discovery and improvement, it’s all about timelines. For those who can shave off one or two years from the ten to 13 years it takes to get a medication to a CVS close to you, then we will dramatically alter the economics of drug discovery and improvement.

Shervin Khodabandeh: So going from the 5,000, which within the outdated world I assume would’ve needed to be examined by trial and error, to the five hundred, you’re limiting the universe that then goes to be developed and examined. Is it attainable that in that means of excluding the opposite 4,500, you would possibly throw away some good candidates?

I’m curious how the educational occurs right here, as a result of in most different AI programs I’m near, there’s a reality knowledge the place the algorithm learns from.

Frank Nestle: Yeah, what’s the reality?

Shervin Khodabandeh: Proper. For those who’re throwing away buildings that the engine plus the chemist would possibly suppose isn’t going to even work, are you presumably throwing away some doubtlessly unborn good drug candidates that by no means noticed the sunshine of day?

Frank Nestle: Yeah, that’s a very good level, and it additionally is dependent upon the totally different steps of the worth chain. Once we make a medication, we’re testing totally different attributes. For instance, only one attribute is perhaps very potent binding. The subsequent attribute is perhaps extremely particular binding, not hitting different targets. The subsequent attribute is perhaps, it’s secure; it doesn’t, for instance, result in cardiac malfunctions or different uncomfortable side effects within the liver. The subsequent attribute is perhaps, a molecule is nicely absorbed by the intestine. So the following attribute is distribution within the physique, into the organ system you want.

So all of those totally different attributes are optimized by our devoted fashions, and these fashions are educated. I’ll provide you with a selected instance. We use what’s referred to as Caco cell strains to imitate what intestine absorption appears to be like like, and we mimic that by understanding how nicely the cell line is absorbing, or taking over, a molecule. So we’re working hundreds of those Caco cell strains on a regular basis, and we’re then learning — the reality is how nicely the molecule is absorbed in that Caco cell line, and that provides us then a excessive rating of a molecule. And the system then learns from this iterative means of what a very good Caco cell line uptake molecule is. And that finally then offers us a greater and higher mannequin. So this mannequin can be then one of many many alternative fashions we might use to foretell these 500 we might finally check. Does that make sense?

Shervin Khodabandeh: Sure. Sure, it does, as a result of what you’re saying is, the molecule you would choose has to have totally different attributes. And these attributes are real-world-tested on different molecules that may look just like these molecules.

Frank Nestle: Precisely. Sure.

Shervin Khodabandeh: So some sort of a clustering, or molecules that appear to be this, or have these sorts of chains, or no matter, or these sorts of ligands, work higher. And in order that’s the way you’re infusing that studying into the algorithm, if I perceive.

Frank Nestle: Precisely.

Sam Ransbotham: I feel what’s vital there’s, I feel we have been responsible of slipping down into this binary classification factor, the place it’s both a very good drug or a foul drug, whereas you’ve bought lots of of various attributes you’re , and you may play with every a type of, and you may check on every a type of in order that it’s not simply that crude up/down, sure or no, that we have been sort of crudely pondering earlier than.

Frank Nestle: We name it an optimization course of. It’s a high-dimensional optimization course of, the place usually, when you then take away a sure molecular piece of the molecule, you would possibly enhance the uptake, however you would possibly lower the efficiency. So it’s really a give and take. And that is what chemists are excellent about, and that is the place they take pleasure in — as a result of it’s such a high-dimensional house, and we’ve got so many knowledge out there — to have that associate, which is AI, to inform them what the AI thinks.

Shervin Khodabandeh: This has been fascinating, Frank. So, we’re going to transition to a section, which is a sequence of rapid-fire questions. So I’m simply going to ask you 5 questions. You in all probability don’t know them but. And so simply give us the primary reply that involves your thoughts. Are you prepared for this?

Frank Nestle: Certain, positive. Completely.

Shervin Khodabandeh: Inform us your proudest AI second.

Frank Nestle: The proudest AI second was after we, for the primary time, may annotate single-cell immunology cell fades, with our in-house-developed AI algorithm at Sanofi.

Shervin Khodabandeh: That’s improbable. When was that?

Frank Nestle: That was 2018.

Shervin Khodabandeh: What worries you about AI?

Frank Nestle: What worries me about AI is that folks don’t perceive what AI is. They at all times take into consideration synthetic basic intelligence as like machines changing people. That’s in no way what I’m seeing. For those who see the place self-driving automobiles are in the mean time, there’s loads to be solved till you get even near it, however what AI actually is, is for very particular questions, with a very good knowledge set, excessive computational energy, and a very good algorithm, to unravel issues a human mind couldn’t do. And these little, small contributions of AI are making all of the distinction, actually by way of what I’m seeing within the R&D worth chain.

Shervin Khodabandeh: Your favourite exercise that entails no expertise.

Frank Nestle: Biking.

Shervin Khodabandeh: The primary profession you wished in childhood.

Frank Nestle: I at all times wished to go down the route of being a author and a director — really, a theater director. After which I used to be studying Albert Camus, La Peste [The Plague], and I used to be learning philosophy and literature. And there’s an individual, Dr. Rieux, who’s combating the pest. It’s very well timed, with the pandemic. And he’s combating this l’absurdité, the absurd existence, by doing good, by serving to.

And I’m not as philosophical, however once I entered drugs, it was only a manner of doing good and doing one thing helpful. And now we do it at scale by hopefully discovering the following drugs, reworking sufferers’ lives.

Shervin Khodabandeh: It’s a terrific antithesis to Albert Camus.

Frank Nestle: Yeah, precisely.

Shervin Khodabandeh: And your biggest want for AI sooner or later.

Frank Nestle: Make it extra explainable, each on the stage of getting it out of the black-box state of affairs, make it explainable by way of understanding what it does, but additionally clarify it to folks in order that there’s not this misunderstanding of AI. Lots of people simply mission their fears or their misunderstanding on that unlucky small acronym. And as soon as you employ it in sure contexts, it’s really transformative.

Shervin Khodabandeh: Thanks, Frank. This has been extremely insightful, and I’m positive, very beneficial for all of our listeners. Thanks a lot for this.

Frank Nestle: Thanks for having me.

Sam Ransbotham: Yeah, nice to speak with you. Subsequent time, Shervin and I discuss with Stéphane Lannuzel, Magnificence Tech program director at L’Oréal. I’m at all times up for a very good episode about cosmetics. Please be a part of us.

Allison Ryder: Thanks for listening to Me, Myself, and AI. We consider, such as you, that the dialog about AI implementation doesn’t begin and cease with this podcast. That’s why we’ve created a bunch on LinkedIn particularly for leaders such as you. It’s referred to as AI for Leaders, and when you be a part of us, you’ll be able to chat with present creators and hosts, ask your individual questions, share your insights, and acquire entry to beneficial assets about AI implementation from MIT SMR and BCG. You possibly can entry it by visiting mitsmr.com/AIforLeaders. We’ll put that hyperlink within the present notes, and we hope to see you there.