An AI Awakening With Profluent Bio's Hilary Eaton, Ph.D.

The company's not at least not right now developing a pipeline , but rather it's leveraging generative AI to do all sorts of useful things like reading and writing therapeutic proteins from scratch and making accessible to the biotech community an open-sourced AI-generated gene editor called OpenCRISPR .

On today's show , profluence , dr Hilary Eaton is with me and , while I'm not going to let her take me too far into the AI rabbit hole , we are going to learn a bit about how it works and , importantly , where and when it doesn't work , its applications in biotech , why ProFluent is doing what it's doing much of it for free , I might add and what's inspired

Hillary's life science career path , both personally and professionally . Hillary , welcome to the business of biotech .

Hilary , you got to bring that . You got to bring your A game . But I want to start out with a little bit about where you came from and what got you into the space in the first place . So a little bit of background . Hillary earned her PhD in molecular cancer biology from Duke .

She did her postdoc at Harvard and then , Hillary , you got seemingly like looking at it on paper , you got directly into licensing and business development deals . So that's where my curiosity gets piqued a little bit . What motivated this ? You know , what looked like a scientific or therapeutic sort of development driven path .

What motivated that switch early on into licensing and business development ?

Science was in my family , so I grew up watching my dad . He was a career scientist , a chemist advisor for his entire life , and he headed the team that developed Lipitor , their famous cholesterol lowering drug , and I somehow got the , you know , very mistaken idea that all scientists just work on drugs that millions of people take .

And that sounded great to me , right . That sounded like on drugs that millions of people take and that sounded great to me , right . That sounded like a really direct path of helping people .

And , as a side note , I think now that I've become a parent myself , this really does make me think carefully about sharing both my successes but also my failures with my kids , so that they don't get sort of like a skewed view that everything in life works out . And and in fact , when I had a follow-up conversation with my dad .

You know I learned a lot about the many projects he worked on that were not Lipitor , that did not make it to the clinic , and it was a very . It was a very enlightening conversation about you know , the ups and downs of a of a career that on paper can look like one very major success .

My daughter's about to go into college , my son's going into a second year of college and they're , you know , in that formative stage and their perception you don't really often think about what their perception of of what you do and how you do it might , how that might influence them . Like you just looked at dad as like the Lipitor making hero , right .

So , you know , maybe halfway through my PhD and that sort of like , you've gotten through the prelim , you're now working in a lab and I started realizing that a lot of the basic science projects that are carried out in academic labs are never going to directly affect a patient's life .

They're often very crucial steps to understanding the underlying mechanisms or pathways that are affected by disease , but they're you know , they're many steps away from a medicine that's actually going to affect lives .

And so during my postdoc at Harvard , I started to search for a more direct way to help people , and I'm a big fan of thinking about sort of inflection points and also pondering on how sometimes what can be a pivotal point for you , the other portion doesn't even realize that they said something that affected you and I come back to .

There was a lunch and learn seminar with the then chief scientific officer of Dana-Farber Cancer Institute . His name was Barrett Rollins and he probably doesn't even remember this , but again , it made a huge impression on me at the time . And he was asked by somebody how did he become CSO ?

Like what was the transition from being a PI , being a lab head , over to a more administrative managerial role ? And in the course of his answer to that question he mentioned what was in know was , in principle , a fairly simple exercise . It sounded like something that like a career counselor would give you to do back in high school .

And he said you know , make a Venn diagram and on the one side , have it be the things that you're good at versus the things that you're not and be honest , and then have the other side be a list of things you enjoy doing versus the types of things that you simply tolerate or maybe in fact hate .

Um , and then he said if you in your current role are not at that intersection between things you are good at and things you like doing , that does not mean that you are a failure at the thing you are trying to be . It means that your skill set and maybe your enthusiasm would probably be better fit for a different role .

And that really struck me , and so I went home that night and made a Venn diagram and first of all , it was a lot harder than I thought , like getting honest with yourself about stuff that you aren't good at or that you don't like doing . That can sometimes be a little depressing to to see you know .

Kind of look at the current job you're doing and the list of things you don't enjoy is much bigger than the list of things you do enjoy .

But it gave me a hint about what to look for next , and that was exploring the business side of science , being able to still leverage the skills I'd learned throughout graduate school and my postdoc , but sort of applying , you know , some of my strengths in communication and in problem solving towards how could you make medicines and get them successfully to patients .

And so I started sort of on , you could say , the light side of the dark side by dipping my toes into technology transfer at Dana-Farber , and that was really a crash course in many things that I had expected .

I got to learn about agreement negotiation , intellectual property , budgeting , alliance management all things that I had never touched on as part of my scientific training but surprisingly , I also got a crash as part of my scientific training , but surprisingly I also got a crash course in expanding my scientific horizons .

You know you work on these niche projects during your scientific training and so much interesting science goes on beyond your project or your department at a research institute and so I got , you know , honestly , a broader picture , I think , of the science being done at Dana-Farber during my time in tech transfer than I had as a postdoc with my head down working at

the bench . And it was , you know , I found myself every night back on PubMed , you know , trying to learn about new areas of science that I hadn't been trained in . Like there was this new field called , you know , oncology .

I knew nothing about it but I got assigned a portfolio for a couple of lab heads that worked in that space and it was absolutely fascinating to me to get to see kind of the best things , the most cutting edge things that were going on , and then to try and figure out which one of those things had legs to become a company or to be licensed to a pharma to

become a medicine , and that was my introduction to business development .

Maybe the balance is a little bit different , or was it ? I'm just curious about , like , how that experience influenced .

It's that , like when I look at people , it's you can see when there's a sparkle in their eye , you can see when they're enjoying what they're doing in that moment , and you see times when that spark kind of dims and it's just the rote grind and I don't know if there's ever going to be a job where your eyes are sparkling a hundred percent of the time , cause

that might just be complete burnout or or just , you know , irrealistic . But I definitely noticed that there were a lot of things um that spark to my eye , this idea of you know doing , for example , a deal negotiation . Ultimately it's framed as you're on opposite sides of the table right and trying to get the best deal for your side of the table .

But that's not really what it's all about . What you're trying to do is get at whatever science , whatever drug there is to be made there , and both of you want that and each of you have some very real concerns about things that could go awry on the way to building that drug .

And so figuring out sort of what the actual concerns are under the surface and then finding creative ways to compromise about those , that to me is fun . It's a fun puzzle to work out .

And , speaking of deal-making , I think next , stop on your career . You're doing big deals at Editas , right , was that ? Was that next after Dana Farber ?

So tell us a little bit about that story .

So you know , as I was looking at the Editas role in the background , my husband and I had been trying to have kids for some years without any success and we ended up doing IVF . And so , ironically , I was actually waking up from having my eggs harvested when I got the call and I was like , yeah , I'm taking , I'm making this decision .

And so , you know , while I was at a task , had two little IVF miracle babies and was just so grateful that the career that I had embarked upon and science was able to sort of make a dream come true for us .

And then , over the course , you know it's so busy , like two parents working full-time in the biotech , you know two young babies , and you know it's just kind of like all a blur .

But you know , at some point along the way we kind of experienced one of those like worst moments that you can have as a parent , where all of a sudden you realize that your kid is missing developmental milestones and , like you know , when it's your first kid , you don't really have much to compare it to , right , you don't have benchmarks out there , and so it

was kind of like , you know , yeah , she's taken a while to sit up , but , like I don't know , there's a range . The pediatrician every time tells you , oh , kids are a range , it's a whole spectrum , right , but it was more prominent in our second daughter , and at this point actually I think I had transitioned to Vorbio and she had some hearing loss at birth .

But it's so hard with babies to figure out how profound certain things are . You know , they don't exactly cooperate on the exam table with various tests , and so it wasn't until she was nine or 10 months that we really realized how significant the hearing loss was , and there were a variety of other symptoms that kept popping up .

But the biggest thing is , you know , we were in it at one of the hearing sessions and somebody says , you know , oh , could she sit by herself in the chair without you holding her ? And I said , oh , no , she can't sit up by herself . And they said , really , she's 10 months now , that's .

You know , she should be able to sit , and so that led to being introduced to early intervention , which the state of Massachusetts does when you have a child with two or more developmental delays , and a whole host of additional specialists to look into why she was missing these gross motor and these fine motor milestones .

Know what it came down to in the end is that she was so hyper mobile , so flexible , that it was hard for her to build strength and figure out how to do some of these crucial things like sitting up or pulling to stand or crawling .

And I can remember , you know a point where I was in I think it was an orthopedic office and Olivia that point was , I think , maybe around 26 months and not walking , so so significantly delayed and just feeling extremely frustrated because we had gone through sort of a laundry list of different you know pediatric specialties and nobody was giving us answers .

And sitting in this office and the doctor turns and looks at me and says mom , are you bendy ? And I said I mean yes , what does this have to do with this conversation ? She was like do you have party tricks ? Like you know , you double jointed . Can you bend joints backwards ? And I said yeah , and she was like I know what's going on here .

She's like there's this thing , there's , there's these host of connective tissue disorders where the most obvious symptom is this hypermobility . And she's like , and I think what we've got going on here is that this little girl , you know , is just so , so bendy without the corresponding strength .

Because how would she have built that up without the corresponding strength ? Because how would she have built that up , right ? I think that she needs , you know , some braces . She needs some braces on her ankles , and just stabilizing some of those joints is going to really help with the immediate symptoms .

And it was just that moment of could this really be it ? Do we have some light donning ? And so they fitted her with some custom orthotics and we kind of thought , yeah , try sticking your two year old ankle braces . There's no way she'll take it . You know this is going to be a grand tantrum .

And instead we had that moment in the office where they put the ankle braces on her and she turned and looked at me and then she pulled up to stand like I was just . I was just waiting for this , you know , a little support .

And so I think , you know , we got phenomenally lucky to be in Boston and to have some amazing physicians in our network that were able to get us there , you know . But but one of the issues here is that there aren't , just as there aren't treatments , there are also not diagnostics for many of these things .

So the this particular disorder is called hypermobile Ehlers-Danlos syndrome , or HEDS for short , and for this particular sub variant of disease there are no known genetic variants . So it's not . You can't just send a blood sample in and have a doctor say , yes , you have the mutation , or no , you don't .

Instead , they're looking down sort of you know , a bucket list of clinical symptoms and if you meet enough of the criteria , then you're deemed to have the disease .

And the thing is , my girls at the time , two and four , they were too young to have had some of the life experiences that would have led to some of the symptoms , like repeated dislocations , slow wound healing .

And so the doctors came to me and said you know , mom , an alternative approach here would be if you got officially diagnosed , that would allow us to then diagnose your children .

And so I found myself in a doctor's office at I don't know 37 or something you know , being asked all these questions and all of of a sudden , so many different symptoms or weird medical things that had happened to me over the course of 37 years were drawn together into a diagnosis .

And I still can remember , you know , the doctor at the near the end , the geneticist at the end of the appointment , said , you know , heartwarming that you're , you're doing this , you know , in order to get your kids , you know , diagnosed and on the path you know to , to ameliorating their symptoms .

He was like also , you've just been living with this for several decades , you're , you're kind of screwed up health-wise and you need doctors . Yeah , you need to go see specialists for some of these different conditions .

And I just thought it was too funny that progression sort of like to diagnosis , that like symptoms that I had experienced somehow weren't worth it to me or didn't didn't rise above the radar to maybe I should go see a doctor about this . But when it was my kids , the mama bear comes out and you're like , okay , let's get this solved Right .

So yeah , that was , that was the path to getting diagnosed .

It's actually led to our whole extended family being tested , because , you know , there's the part of this journey that's getting an answer for yourself . There's also a part of this journey that's how can you help the broader community .

So when somebody shows up in a clinic and there's no family history of any of these symptoms , it's very hard then to look for what in their DNA might be causing these symptoms .

Sure , when you have a large extended family with folks who are quote unquote normal and folks who are less normal , you can start finding patterns where you can say , okay , you know , maybe we have 75,000 snips or random mutations in our genome , but there's this , only this one snip . That's in the affected females in this family .

So it , you know , we saw it as both answers for ourselves . But , just as science my husband and I are both scientists , my dad's a scientist we had this . You know , immense curiosity about what is causing this , and can we add to the knowledge in the field ?

I mean , during my time at Dana-Farber , one of the things that came up is that they treat so many patients every year and there's data from treatment outcomes , there's data on symptoms , there's also data in the form of actual patient samples , like biospecimens , and you learn a lot from seeing how somebody's tissue changes over the course of disease progression and

then treatment .

And so one of the things that we worked on at Dana-Farber is making sure that when you're writing up IRB consents , irb protocols to enroll patients in clinical trials , that you can really explain to them how valuable their data is and how it might be used in the development of new medicines and what precautions you take to keep the right privacy around that data .

And then to find myself sitting on the patient side of that seat you know , reading patient consents that I might've had a hand in writing . You know it hits hard and there can be that instinct of like no , no , my genes are mine , my information is my own private , and I respect people very much who who make that decision .

I think maybe you know somewhat uniquely because of the science background . Part of me said , yeah , I'll put it all on the table . If somebody can learn something from my cells , my DNA , that means there's a mom down the line whose baby is diagnosed at birth and who learns some of these you know developmental things on a regular schedule .

I will have contributed in you know in a meaningful way that has you know nothing to do with like a personal achievement at all . It's just something that's like inherent in my body .

You know , you have connective tissue in literally every place in your body , so there are a whole range of things that can go wrong with this disease , and it used to be that they would only pick up patients fairly late in life .

I think the last quote that I heard from the patient community is that the average time to diagnosis for Ehlers-Danlos is 14 years . So you're often dealing with folks in sort of young to mid adulthood . Now Boston Children's Hospital is seeing a lot of much younger kids coming in and they don't really know what disease progression is going to look like .

They don't know how early certain symptoms are going to are going to pop up , and so you know , we just made sure that we got , you know , the right team in place , um , with regular check-in schedules and and it's you know thing , we'll , we'll solve problems .

It can be a little bit like whack-a-mole , because you know you think you've gotten to a good state and then something new pops up . But I think I'm just so incredibly lucky that that I have this network of of friends and colleagues and former colleagues that I can draw on when some of these new things pop up , to figure out .

You know how to frame the approach . It's , you know , something happens , so figure out how to solve it . Figure out who the people are to talk to . Um , they're just they're doing wonderfully . They're both in school now .

They're they're just really thriving and in some ways you know as much as I was , you know , sad to find out that I have a disease that I had passed on to them . You know my therapist , my executive coach , all my friends , my therapist , my executive coach , all my friends .

The point to drive home is that who better to parent a child with rare disease than a mom who already knows what it's like to live ? There's nobody better to advocate for them .

That does not define me , does it limit me , perhaps in certain ways , but you know , it's interesting to figure out how you can , sort of like find hacks or band-aids or approaches that allow you to to ambitiously go after your career goals while still dealing with some of these things that are happening in the background .

Sometimes I almost think that maybe that in and of itself is what led to me having certain strengths in my career in business development , because I'm not afraid to go out there and advocate for things and ask the hard questions and try to come up with creative solutions .

I had a couple of postdoc friends who were using it for screening , but I had no idea that it was going to become sort of a next frontier in genetic medicines . And as I looked , you know at beyond Editas again , these mentors were saying , okay , you've had , you know , an incredibly successful stint at you know , one of the pioneering editing companies .

Do something else , you know , go into a different area of technology or a different area of disease of technology or a different area of disease .

But for me , when I looked at the field , there is so many good science , there's so many other good things , particularly around the Boston area , that you could go for , but I do not see any other path to alleviating or even curing rare genetic disease that's more promising than gene editing in its many forms , and so I've stayed put .

That's definitely been a theme for me staying at companies that are exploring with different generations of gene editing systems , um and yeah , and in many ways it kind of feels like that is the perfect intersection of life experiences for me that I'm trained as a scientist and , fundamentally , gene editing is not an easy or a simple process .

It is a very intricate system , complex biology in business development . I really never know who I'm going to have to explain or interpret to .

I could be dealing with a Nobel Prize winning scientist who wants all the details , or I could be dealing with somebody who has more of a finance background and needs it to be explained in a simpler manner that they'll be able to deal with . And so you've got that part of it .

And then , on the deal negotiation side , one of the things that's been really interesting then is getting to see , as this , you know , field of gene editing has started blossoming over the last decade or so .

You know , seeing what it takes to have good ideas and good strategy and how some of those things pan out , like what decisions you make with the data you have in hand at the time versus what decisions you know you might've made if you'd had 20 , 20 hindsight . And then , finally , that patient perspective right it is .

You know , I remember the , the , the clinical indication selection exercise at Editas and it's very patient driven , it's very patient focused . But that patient is also other . That patient is sort of you know , cut and dry facts that are on a piece of paper . What does the disease progression look like ? What are the symptoms ? What's the underlying genetic mutation ?

Can our technology fix that mutation ? And once you get a diagnosis , the psychology of that exercise is so very different because it becomes you know that that could be my daughter right ? How are we approaching this safely ? Is this you know ? Can we with with the way that clinical trials need to be set up ?

Are we going to be able to intervene early enough in that child's life to be set up ? Are we going to be able to intervene early enough in that child's life to have meaningful clinical benefit ? It just , you know it changes the equation , and so I think you know one of the things .

There are so many rare diseases and so many patients out there who are looking for answers that I really don't think there can be . And it also , then , has changed in certain ways the way I view the business of medicine .

Right , it's phenomenal to see , you know , the Casjebi , the first approved medicine , crispr medicine for sickle cell , that CRISPR therapeutics and Vertex , you know , brought to the clinic , seeing folks get cured and no longer having these horrible viso-occlusive episodes with , you know , pain and blocked blood vessels . That is amazing to see .

And those drugs are priced accordingly . I mean , you know hundreds of millions of dollars of research go into . You know , every one successful drug has 99 failures sitting in the background .

And yet , when you look around the world , you know the patient population of folks with rare disease is not limited to , you know , rich people in developed countries that have great insurance . It spans the gamut . Nobody's spared Right . And so you start thinking about . You know we are very early in the arc of a new technology .

Things are usually most expensive at the beginning because they're so new and you see , you know , as things become more common and commoditized you do see prices come down . But you know , it really did start getting me thinking about , like , how do we Jennifer Doudna and her Innovative Genome Institute right this idea of democratization ?

How do you take such a powerful technology that has applications that go way beyond human therapeutics , when you think about sustainable crops , climate change , and how do you responsibly use it for some of these things ? How do you use it for the betterment of society and how do you do it in a way that's approachable for as many patients as possible ?

I talk to a lot of physician scientists turned biotech founders and they tell me well , you know , I was administering the thing and the things and then I made the decision to get into biotech or biopharma because I wanted to make the thing right that I was administering .

So here you are you're working for companies that are making the things , a few of them in a row , and doing well I'll add , great business development career with these companies Doing what a lot of people aspire to do making the thing , developing the thing . And then what ?

Last year , early this year , you decide now I want to go do something else , and it's not with a company who's making a thing per se in terms of a therapeutic . You were talking about these bigger problems . You know these bigger challenges that your current company , profluent , is attempting to play a key piece in the puzzle right of solving .

So tell us about that . What prompted and how was , like , what was going on in your mind ?

What did you struggle with when you said you know what , like , I think I'm going to move away from working in BD for therapeutic development companies in the gene editing space , and I'm going to move away from working in BD for therapeutic development companies in the gene editing space and I'm going to jump on this yeah yeah yeah , and I don't want to call it

a bandwagon . I mean , yeah , I'm fully on board that bandwagon and I don't even want to call it nascent , like it's here . Right , it's happening , but still big transition for you . Big decision , right , it's happening , so , but but it's still big transition for you .

Big decision , right Walk away from therapeutics development and and uh , and take on a risky role with a startup company that you know doesn't really have a a product per se . When you joined , like what was going on there ?

the developing gene editing field as a whole . One of the things that I was struck by is a pattern of needing a new approach to protein engineering . So why would we need a new approach ? What wasn't working or what isn't working ?

So in many cases , crispr being one of them , the starting point for a therapeutic is a protein sequence that has essentially been copied and pasted from nature , and often it's by random chance or discovery .

It's not that somebody's trying to find that particular thing in nature , it's that it's , you know , somebody's working in bacteria and then discovers this thing that does something , and they start thinking , huh , how could that be applied ? Could that be applied to human medicine ?

Now the tricky thing is that you know a protein that evolved in nature to carry out a certain function might not be the perfectly designed solution for a given human use case that you're looking for . And so you know , if we want to adjust the function , how do we go about doing that ? We need to tweak it slightly . How do we go about doing that ?

And so there are sort of you know , different approaches or different eras of protein engineering . It's not a new thing , it's been going on for decades , right ? So one is sort of the manual route where you could carry out random mutagenesis or directed evolution and , you know , trying to guide the protein towards those criteria you're looking for .

This is incredibly time and resource intensive . It often is sort of you know again , random mutagenesis does not really imply that you have much of a hypothesis . It's your sort of like randomly know . Again , random mutagenesis does not really imply that you have much of a hypothesis .

It's your sort of like randomly tinkering with different areas of the protein to see if you then get results that you want , and often what you end up getting is incremental improvements at best .

More often you break the protein because you tinker with a fundamental area of it that's necessary for function , and so you know then , maybe moving on to the next era or sort of a different approach , we've got folks who do metagenomic mining . What does that mean ? Well , they look at . You know , I lifted this particular sequence from this particular organism .

It's almost right , but it doesn't do these other things we want it to . So I just need to broaden my search . I need to look at more organisms and in the diversity of organisms eventually I'll find something that does have my list of criteria . And certainly some folks have had success .

You know , looking into , you know volcanoes and ice flows for more genetic diversity . But I would still argue that that is a fairly needle in a haystack approach and it often results in functional trade-offs . And so what do I mean by that ?

So when you look at natural evolutionary processes , often the selection pressure that's being put , that leads to sequence changes , is focused on one particular attribute of that protein , of that enzyme . So , for example , perhaps the host organism is small , so the enzyme itself needs to be smaller .

It just isn't going to fit , it's not going to be able to do its job . But in the process of evolving and achieving that smaller size , it's possible , I'd argue even probable that other aspects of that protein's function are going to be altered . And so change is kind of like a double-edged sword in this scenario it's .

You know , you achieve one goal but you sacrifice other features that weren't critical for fixing the current problem in that original organism or survival of that organism , but they are key for the therapeutic problem that you want to solve . And so , you know , in comes profluent and generative AI , which , I'll be frank , I knew nothing about .

I'm a huge sci-fi aficionado , but I knew nothing about real life AI before being introduced to profluent .

And you know , one of my first thoughts is this is the next era of protein design , where we're moving away from sort of accidental discovery or sort of you know , tinkering around in a manual way , and we're really moving to intentional , bespoke design aided by artificial intelligence . And you know , they have these large language models .

And those large language models they're not limited by natural sequences , they don't involve manual labor . There's no functional trade-off .

When you look at the combinatorial space of a single protein , it's greater than the number of atoms in the world , and so there's no way that even a team of the best scientists can sit down and explore all the possible sequences , right . But generative AI can do that in a matter of minutes , right .

It helps you narrow the funnel of what could be possible places evolution hasn't had a chance to explore and pull the gems out , and so , rather than sort of starting with a molecule and then trying to kind of like jam it into a therapeutic use case that it may not be quite perfect for but it's sort of good enough , we're really trying to kind of flip that

paradigm for drug development upside down . Start with the problem what is the problem you're trying to fix ? And then if you had to imagine the perfect drug that would fix exactly that problem , what characteristics would it have ? And once you have that list of criteria that you think define the ideal system , then we task the AI with building exactly that thing .

And that just blew my mind , that could that even work right ?

Okay , this , this description that you just uh , that you just illustrated , um , start with the problem and then ask what are the attributes of , of the solution to that problem , and then go out there and use the data . It seems to me , in that situation , the data , at least at this point in time , might be the weak link .

Like to query to apply these models and make impact .

It's definitely sort of like a fire hose right At the beginning of a new job picking stuff up , and we're lucky enough we have a board member named Fraser Kelton , who was the head of the former head of product at OpenAI for ChatGPT and DALI , and then ProFluence own CEO and co-founder , ali Madani , who was one of the leading researchers at Salesforce that did

all of this foundational work on applying large language models for protein generation , and so they've definitely helped bring things down to my level in terms of understanding what are large language models and how did it go from an area ?

If you look back five , 10 years ago , your iPhone model I don't know two , three couldn't finish a sentence for you Like it's , it's , it's predictions , right , we're , we're often hilarious . I would sometimes just like take the prediction because it was so hilariously wrong , right ?

So how did we get from that world to a point where we have something like chat , cpt where you can say write me an essay in Shakespearean style on soccer and it needs to be two pages long and it can just do that . Right ? That it seems like an incredible advancement . How did we get there ?

And , as they describe it , the two key ingredients you know one really is fundamentally advancements in AI . There's something called the transformer algorithm that apparently had a profound effect on models' ability to learn and generate useful results . But the second key ingredient is what you are alluding to .

It's data , it's the scale of data , it's the quality of the data , it's the bias or lack of bias in the data . And so you know , when you think of , you know how things like ChatGBT were able to have success .

You know there are millions , if not billions , of examples of text floating around on the internet , in you know , books , all these different things , right . And so that's a rich abundance for the model to scrape and learn from and learn .

You know , just by like , when you think of a young child learning language , right , some of it's intentional teaching , but a lot of it's just passive absorption . They hear things , they learn grammar unconsciously , and that's what these models do . And so now , if we turn to the world of protein discovery , you know that same abundance is mirrored in nature .

We have , the world has provided us with this abundance of natural protein sequences over the course of evolution , and much of that data is , just like , readily available . It's just out there .

You know , I haven't been around the life sciences space long enough to know that they were jumping on that bandwagon from the outset and saying all the data , all the data , all of it , it needs to go , it needs to be digitized and it needs to become accessible . So that I think that's kind of central to my question .

So I would say that there are ways in which some data doesn't exist and we can get into that , I think later , maybe , when we talk about sort of like negative data when it comes to things that are natural sequences , as opposed to , you know , a drug that somebody is trying to develop , when we think about what exists in natural sequences yes , there are .

There is so much data out there that's digitized . Our current frontier model is being trained on over 80 billion protein sequences that are just out there .

I can't say that they're often perfectly curated or ready to go , so there's a lot of intentional mining , tidying , curation that needs to go when you pull down this information from all of these different sources , both public and proprietary , so that it is apples to apples and your model can actually learn from it .

But there's so much of that that exists out there and that's what we call unsupervised data . This is just sort of like you want to expose your model to as many examples as possible so we can start playing little childlike games over and over trillions of times .

It's think of it like like fill in the blank exercises , right when you know , if I say the girl kicked , you have learned that probably something's going to come after that . It could be the ball , it could be her brother , right ? But there there's a narrower universe of things that could come in . That fill in the blanks .

And if you just started a sentence with the and , so , in the same way , with these protein sequences , the model , as it's exposed to it , it's learning the language , it's learning the syntax and grammar , it's learning what protein domains are responsible for what functions , it's seeing evolutionary dead ends where things didn't work and ultimately , what it comes out

with is the ability to write things from scratch . So I'll be very clear here Our models are not starting with a existing sequence and then performing the AI version of targeted mutagenesis . They are being fed input where you say I have this therapeutic problem , to solve it , I need an enzyme with these qualities .

And it says okay , let's start with amino acid one . If I know what I want the end protein to be capable of doing , what should the next amino acid be ? What should the next amino acid be ? And it generates millions of these possibilities and then ranks them based on its own prediction of how confident it is that that's going to be a successful hit .

And that's where the second type of data comes into play . So you have this unsupervised natural data that you're exposing the model to to train it . Right Now you have supervised lab data .

So there's this sort of like feed forward loop where , if our models have generated certain sequences and they think the model thinks that these are going to be you know , crisper analogs , for example Right Now you have to go test them , right .

So you take the top five and maybe a couple sprinkled around that it doesn't think are going to be as good , and you put them into the wet lab and you run functional assays and you see how they perform in real life .

And now you've gone from like the theoretical , the hypothetical , to like rubber meets the road , like what does this actually look like in practice ? And when you get those results back , that's supervised data , it's labeled data .

So you feed that data positive or negative back into the model and it learns which of its things were successful , which ones weren't , and maybe it sees patterns in terms of why , what made the successes a success and vice versa , right , and so you know . It then learns from that data and I think that's a really interesting point .

That ties into sort of the idea around both is there enough data and what kinds of data are there ? I think both . Is there enough data and what kinds of data are there ? I think , you know , in the field of biotech and medicine , there is an emphasis on successful data . That's what gets you papers , that's what gets you drugs right Publish or perish .

And yet in the field of tech , there is a very different attitude towards quote unquote failure . It's , you know , move fast , break things , learn from them Right , and that view of when something fails , you can learn from it , the models will learn something . So it's not a waste of time , it is , in fact , very .

If your models were only exposed to positive data , that would not be a good thing . They would not be as well-rounded , and so that , for me , was a very interesting psychological shift that you're almost , you know , as intrigued by negative results as you are by positive results .

I sat in on multiple discussions with folks who talked about that Like you're , you're so convincing Hillary , like I'm listening to you and I'm like why would anyone be skeptical ? Can you answer that question for me ? Why ? Why are , why is there so much skepticism ?

I think that's the root's a lot of top level messaging or or miscommunication around what something is and what it isn't . That can lead to a lot of this confusion or skepticism . So , you know , first off , I would say it depends on the exact question that you're trying to answer .

So if you're talking about skeptics who just think that AI in general and or for biologics is just total hype and not actually reality , I would say , like , is AI a magic wand that you can just wave at a difficult therapeutic problem that your best team has been trying to solve for 20 years and it just instantly solved ? No , it's not .

So if that's the type of hype we're talking about , then I would probably agree right , is AI a highly complimentary tool that can be used to augment the pace of innovation in science and in therapeutics ? I think yes , and the key that I see there to it having a realistic impact , and it is having . That , now , to be clear , is collaboration .

It's not some machine learning scientists locked away in their cubicle coding and then coming out and saying , aha , I've solved cancer .

It's this tight collaboration of folks on the AI , machine learning side , working with biologists , manufacturers , clinicians , in whatever the biological vertical is , and saying , look , you know the problem , you have intimate knowledge of how to design a therapy , what a clinical trial looks like .

Tell me things , tell me what your ideal attributes are , and I will translate that into what the model needs to be exposed to and what types of output that it's spitting out .

And when you have that tight collaboration , particularly when you're talking , you know , in our case , about CRISPR and AI those are two , you know , hyped up , sexy , inflammatory topics , right , but then disruption plus disruption . You know where's that going to take us . I think that's phenomenal .

Maybe , building on the second point , then you also you have folks who who think that it's powerful but don't really understand how it works , and so then have that sort of level of skepticism . Maybe it's not skepticism , maybe it's just not understanding how to apply it .

Right , excited , but don't know how to apply it and I think my executive coach would say reframe your skepticism as curiosity , right ?

You know it's really a two-way street here and I don't know where the burden lies , but I would say there's some responsibility for leaders in AI to educate both society , but specifically the broader , you know , scientific community , about exactly what AI is and isn't .

It's one of those things where when somebody tells you that you can just go boop , boop , boop on a computer and you're going to get some , you know , perfect proteins , potentially perfect proteins spat out , there's that innate sort of like oh , hell , no , like . You know what I mean .

I've been doing this for years and I have spent , you know time , blood sweat , trying to make these therapeutics . There's not . How is it possible that , like a robot can do this , right , right , and then you see it in action and if you're able to get over that kind of sense of like jealousy or you know sort of like , how is this possible ?

This invalidates everything I've been working on personally . Right , and see the promise of where that could be taken with you know tight collaboration and and and supervision and guidance , you start to see , oh , my goodness , we could take some of the most naughty problems that folks have been , because what is the ultimate goal .

The ultimate goal is not how you get there , it's just to get there , it's to have the medicine , and if AI can help with that , I don't see why we should treat it any differently than any other tool that we have brought into the toolkit over the past decades of pharma .

You're talking about , like you know , you cycle these outputs back to the wet lab and then that wet lab experience creates more data , perhaps negative , to get back to that negative data , perhaps negative , and you feed that positive and negative data back into the machine .

This iterative , iterative process , and I want to one make sure , make make sure I'm correct here .

It takes a look at what worked , what didn't , and it then iterates on those parent sequences and creates child sequences that usually are better right , because it's learned valuable things and that iteration process back and forth you know most of the problems that we've worked on to date .

We've needed about two to three cycles back and forth , and so it's just , it's incredibly fast to be able to get sort of that validation and then immediately have better options to try out .

But I'm curious about your perspective as a business development person like you're the business development person who's like out there engaging the industry with this uh solution set that profluent is working on . What are you seeing in terms of industry appetite ?

And what you're seeing more at JP Morgan and many other venues over the past 12 months has been does AI have a place in earlier stage in discovery and development , research , and a lot of examples that are proving out . Yes , it fundamentally does , across a wide array of verticals .

Right , I think some of the first comers were more in the antibody biologic space , and then you see folks doing small molecule research with AI and then other types of proteins , and so I mean it's here and I think you know , in terms of the industry appetite and adoption for it , that appetite's definitely there .

From my perspective , you know , on the BD side of the startup , there's a distinct bifurcation between sort of like flavors of big pharma . So on the one side , I see certain companies that have really invested enough who know AI and they don't .

It's not always folks being like I'm going to go get a you know , a second PhD in machine learning , but they bring in the right people to sort of augment their capabilities and those are the folks who can do deep diligence . Right , because nowadays there are hundreds of companies who plaster sort of AI , machine learning , deep learning on their website .

When you look under the hood , you get a sense of you know how cutting edge and innovative are they in the field of AI , when something's new and people don't know much about it , the glitz and the glam can be very easy right To sort of hide things .

But , like you know , is a company using , say , somebody else's open source code and recycling it for their own purpose ? Or have they made AI advancements in and of themselves ? Do they have platform innovation on the side of AI and not just on the side of drug development ?

Right , and so that's kind of like you have , you know , that sort of flavor of big pharma that's made the investment , that understands it and that knows where they want to apply it . These are the folks that will come to us saying we have a problem , we have strategic alignment or focus on a particular indication .

We see this particular mutation in this patient group's DNA and none of the tools that we're working with right now can efficiently address that mutation , fix that mutation , and we know that to address it we need something with these exact criteria . Can you build that for us , right ? That's a conversation that's fun .

Um then , perhaps the second bucket is more , uh , the folks that that see that AI is sexy , they see that everybody's talking about it , but they don't understand it . They have some fundamental skepticism or they want sort of you know , a proof of concept , a look see a pilot .

They want to test it out before they invest too heavily , and often they don't know what they want to use it for . And you know , our CEO , ali , is fond of saying why bother using AI for problems that could be solved without AI ? Reserve the big guns for the problems that are intricate and complicated enough that you can't solve them by yourself , right ?

So we don't want proof of concept experiments . We want big , meaty , ambitious moonshots , right ? Because that's where you really test your technology out , right ? That's where you know you figure out the difference between hype and reality .

Want these great big giant projects , these big problems to solve ? How do you like ? I'm curious about give me a little hint as to what the business model looks like there right now ?

I think we looked at sort of the early stages of the company and said what is our strength right now ? We're small , 25 people roundabouts . What we know is protein design . What we have is some amazing machine learning scientists and some amazing editing scientists who really can make that feed forward loop go quickly .

And so why not play to our strengths and why not have a success before you sort of build too ambitiously , right , why not ?

You know Ali's paper from back when he was at Salesforce in nature biotech , showing the first example of somebody could take large language models and write proteins from scratch and then they worked right , that was a moonshot in and of itself .

And to go from the theoretical to the practice and then to start thinking of , you know how are you going to apply this right , right , why not find partners who are the experts in that given field or that given disease , who understand what it takes to take something from that early discovery lead stage and turn it , get it to the clinic on on sort of that

accelerated timeline and we each play to our strengths right .

A lot of you know this is maybe more in sort of like the biotech or the startup world , but I will say that often what happens is when you're fighting for VC money and fighting to stay alive in markets like this , one of the first things that often gets cut is your platform team , because often by then you have a couple of programs in the clinic or heading

to the clinic and you just can't justify keeping a platform team alive if all you're seeing is sort of you know they're putting in tons of hard , amazing work but they're getting kind of these incremental advances in proteins and so in some ways , that's . You know what we are in .

A module is , you know , your protein design team from your wildest dreams right , with capabilities that you know . Again , a lot of companies haven't necessarily invested in the AI infrastructure to be able to do protein design at the level that we can do it .

And so you know as we've been , you know as we debuted , you know , one of the reasons I think , for open sourcing , opencrispr , was this idea of how do you prove that you have what it takes to live up to some of these promises Right , and to be able to show that you know , for the first challenge that we took on publicly right wasn't necessarily a let's go

find some like small , easy enzyme family and then see if we can make you know a competitor to that . No , ali looked for hard problems . I remember some of our earlier earliest conversations and he said tell me about editing , tell me about white space . How , not us trying to build a spy Cas9 competitor or some perfect editor for a given use case .

It was literally a proof of concept of us saying we've trained up a model , we fine-tuned that model . If we now let it run and tell it to make something similar to Cas9 , but with a completely dissimilar sequence , can it do that and do those sequences function ?

You know , cutting DNA is not an easy task and going through that project and coming out with , you know , some top hits that had amazing activity that was absolutely phenomenal , right , and so Open CRISPR itself . That's just one sequence out of millions in our portfolio .

So in terms of business value , the inherent value around one sequence , it's not quite the same . Like I think , if you look back at sort of the earlier days , there's a lot sort of focus on static assets , on well , I own this molecule , I own this sequence right and I will license it to other people who want to make medicines using that molecule .

I think we're maybe moving a little bit beyond that to less of you know what the sequence is , but more of can I build the best thing that will fix this problem ?

And there are maybe hundreds of ways I could go about building different things to fix that problem right , and so then it becomes a little bit more around , sort of those ambitious problems and the products that you're helping somebody design and build and test .

And so we just ultimately made the decision that what we would like OpenCRISPR to achieve is accelerating innovation . You know there are a lot of folks out there who are very interested in solving problems with CRISPR technology , but for a variety of reasons , maybe they're kind of , like you know , a little intimidated by the intellectual property landscape .

Maybe they can't afford the royalties . You know they're looking for other options , and so you know you don't see open sourcing really happen in the world of biotech .

It's very common in tech , right , where it's like you know , new software , I'm going to just give you my raw source code and you're going to break it and you're going to make it better , and and the whole idea behind that is to advance the pace of innovation , right . And so you know we started wondering . Similarly , I thought it was a crazy idea at first .

I'm not going to lie , it's grown on me , this idea of you know that was never going to be per se , our moneymaker . There are tons of enzymes out there that are simple cutters , that are nucleases , and so why not just give this one out to the community and say use it for whatever you want , use it for research , use it to build a therapeutic .

You don't owe us a cent , just go and make good things with it . And you know , what was fascinating to me is I kind of thought you know , what kind of demand are we going to see , what kind of interest are we going to see ?

And you know , from a business development perspective , I think , when I look at you know at the past , for given sequences , you know , depending on your licensing paradigm exclusive , non-exclusive , maybe you see four or five players bidding on something . We got hundreds of license requests in the first week which we were totally unprepared to deal with .

Frankly , right , that type of demand we just hadn't anticipated , and from such diverse sources folks in agriculture , folks in biomanufacturing , folks in therapeutics , big pharma , academia , small startups , right . And it sparked a lot of really interesting conversations that you know . That led beyond OpenCRISPR and more to .

Okay , if this is the proof of concept of what your models can build , now build me this other thing that will solve this other problem I have . And so you know , I , you know , stay tuned . I think you're going to see some interesting things from ProFluent in the next year .

Like what you're putting together there , uh , new as it may be and and agile as the company may be , right , like the direction is not necessarily carved out , uh , in stone , but you're , you're doing well , I mean you've , you've , you've attracted some pretty significant investor dollars . I mean , I think I read somewhere like 44 million .

Like you know , maybe we can't get too crystal clear on what's happening next with ProFluent , but you said from the outset of this conversation I remember you told me when we first spoke a couple weeks ago , because I remember responding to you you said you know , I didn't know what I wanted to be when I grew up .

And I think I responded to you and said I'm 49 years old and I still don't know what I want to be when I grow up . A lot of people are hoping that I don't choose to be a podcast host . Um , but you've been . You've been studied , like we , we , we talked about it Like you . You got into the gene editing space .

You stayed in that therapeutic area for a while and then obviously made this to ProFluent , which is a bit of a departure . But I'm just curious like what , what sort of has driven . You know , beyond the things that we talked about , what , what , what are some of the drivers been that have led you to the point where you are now ?

And what do you see coming down the pike , like how , what's , what's the next sort of a pivot for ?

That being said , in retrospect I think I can pull certain patterns from some of my pivots and moves , and I think you know one of them is definitely the technology that I'm attracted to , sort of disruptive , new , risky technology , because I think that that's ultimately where we're going to see some of the biggest payoffs and it's just so closely connected to my

mission of seeing other patients with rare disease get solutions , of seeing other patients with rare disease get solutions , but even more so than the science , because , again , I mean around the Bay Area , around Cambridge and Boston , you spit and you can find good science . Right , there's a lot of good science , a lot of good technologies .

I think the other major factor has always been the people that you .

You sink , you know , in some ways , more of your life , more of the hours in your day to interacting with other people at your workplace than you might do with with with some of your family some of the time , and you want those to be good people , that you want to be in a room at 2 AM in the morning trying to figure out that next step in drug

development or get that deal done . That's going to lead to the progression of a program and that has been a constant theme I have made .

You know you don't have to be best friends with everybody that you work with , but I have made some lifelong friends at Editas , at Vore , at Tome and now at ProFluent and that is going to think , I think , be a continuing theme .

Working with other like-minded people who are hardworking , who are intelligent and competent and who are just decent human beings that are in it not for the glory , not for the VC dollars , not for the potential ROI , but really , really just want to help people . And this is the way that they've chosen .

And I would imagine that that's going to be where I continue to center interest around opportunities and , you know , at some point I might decide to take the plunge and kind of go full force into patient advocacy . Time will sort of tell . Like I can't wait . You know , I think I should like do one of those like message in a bottle to myself .

Like you know , if it had been 10 years ago , I never would have said this is where I would be sitting , and yet here , I am enjoying every second of it , right , it's like I just I can't wait to see what the next 10 years bring .

I've learned a lot in this conversation and I'm glad that you came on to spend some time with you , and I am also disappointed because I would like to spend the next two hours continuing the conversation , but I just can't do that Anytime .

Yeah , we'll definitely do it again , but I hope you've enjoyed it as well and I'll be in touch to get a round two scheduled .

If you want to hear some different perspectives on AI and biotech , earlier this month our own Tyler Menichiello , hosted a Bioprocess Online live panel discussion exploring how the tech is being applied in process development and manufacturing . That is available now on demand under the Listen Watch tab at bioprocessonlinecom . So go check that out .

Come on back for a fresh episode of the Business of Biotech on Monday and in the meantime , thanks for listening .