Molecular Moments - Stephanie Pasas-Farmer v2.mp3: this mp3 audio file was automatically transcribed by Sonix with the best speech-to-text algorithms. This transcript may contain errors.
Amanda:
Welcome to the Molecular Moments podcast. Welcome to the Molecular Moments Podcast. My name is Amanda Hayes and I'm a scientific officer at Biopolitics. Today I am joined by my fellow scientific officer, Lyn Carmen. And Stefanie passes Farmer. She's the president and founder of Ariadne Software, as well as president and principal consultant of Biodata Solutions. So welcome to the podcast, Stephanie. We're super delighted to have you join us today. And it's always good talking to a fellow Kansan and a University of Kansas Jayhawk. I was wondering, actually, yeah, Rock chalk. Are you from Kansas or what brought you to the state?
Lyn:
That is a great question. No, I'm actually originally from Southern California. I came to grad school here and I used to make jokes about my friends that were from Kansas City, you know, And then that's karma that I married a Kansan in Kansas with the last name of Farmer. So it all worked out well. But no, we we met in grad school. He was in law school, and I was here at getting my PhD. And so we met then and fell in love with Lawrence and moved to Philadelphia for a job but always wanted to come back. And so right before the pandemic, we moved back to be closer to family. And then of course, so no one for three years, but at least we were back here in Lawrence, which is great.
Amanda:
That's great to hear. Stephanie So I actually have a question also about how you came to be where you are right now, but more around your career. So you started as a bioanalytical chemist, correct? Yes, I.
Lyn:
Did. Mm hmm.
Amanda:
How did you get from being a bioanalytical chemist to running a company focused on big data and machine learning?
Lyn:
That is a very strange, circuitous route that I took.
Amanda:
We love circuits here.
Lyn:
So I am not a programmer. I didn't even play one on TV, so I was actually in pharma and I worked at Zero Industry as well. Biomedical chemists, large and small molecule. What you all do day in and day out of biologics. I was gainfully employed and quit and started my consulting firm because I had many different clients. When I was at the CRO asking me to troubleshoot, I thought I would start a lab. That's honestly what I thought I would do. Eight years later, I do not have a lab still, but I love the consulting bit and we have a group of seven consultants now. It was four and one half years ago and I was looking at data and my husband had been reading about A.I. taking over the world and it was Christmas time and I was stressed out because that's that heavy time of year for us in the contract and in filings. And so I had like three screens up. I had all these papers printed out of a clinical CSR trying to review it, and my husband walked past and said, Wouldn't that be great if you had a software that could do that for you? And I thought, well, BLEEP yeah, actually it would be great. I'm tired. My eyes are going square from looking at the screen and tables. Why isn't there something? There should be something and searched and there wasn't anything. And sometimes great ideas come because you don't think you should not be able to do them. And so we built a prototype. Six months later, we had a prototype of some of the auditing software that we have that we actually is a tool for, built as a tool for our consultants. So that's how I got into eyes that I had a problem and ended up meeting someone, collaborator that was a programmer, and we just saw if we could solve it and we did. It's pretty fun.
Amanda:
That is so cool. But you are not a computer engineer, so you had to go out and make contacts in order to get that prototype software to work. How did that go?
Lyn:
Some of it was luck, as I think half of things in life. My husband received a CV because he's an attorney and he has a boutique law firm and someone's CV came in and he not only was a patent attorney, he also had been a programmer and a physicist. And so it was a little bit of a bait and switch. He's like, I don't have a position, but my my wife has this crazy idea, do you want to work together on it? And then we started building the team from there. He has now returned to being an attorney and we have the programming and architecture side of our company still intact. So it was just ended up being at the same time of having that crazy idea. Someone's CV came across my desk that was a good fit and we decided to try it. Part of it is acknowledging when you don't know something, and that could be hard with type-A personalities that we all are here is But we really tried to try to build a team where everyone has a different level of expertise and specialization because if everyone has the same background as me, we're not we're not thinking outside of any box. We're all thinking that was groupthink. And so really, I've been hired. And I pulled from Penn State. And now I pull from cue a lot of local talent. There's a lot of intelligent people that are trying to move in this field independent of domain, just really trying to work in the field. So it's been a lot of fun.
Amanda:
And kind of speaking on that. What was it like taking that risk of starting that new venture, kind of bringing that vision to life? Can you talk a little bit more about that leap that you took?
Lyn:
It was extremely scary at first. It wasn't because it was just very minuscule amounts of investment that I did myself. And eventually, when we got the prototype up and running, the idea of do we take it to the next level or do we just keep it where it is right now is a nice tool for us internally. And that was a big step because I actually have no investors. I did go and look for investors and went down that route and found that for my personal vision it would have muted the innovation, at least at the stage that we were at. We were still trying to figure out what we wanted to do, what we wanted to be. And understandably, investors give you cash and they want return. And that's not always where innovation is flourished the best. So it was a big risk. I actually 100% bootstrapped it and so it was long, sometimes emotional evenings of discussions with my family and close friends that I could trust that would give me honest feedback. But I really had faith in vision, in the vision that we had. And I just knew that partially we had to wait for people to be ready for the tool. I think we were about five or six years ahead of our time because now with the pandemic and people being used to working remote and having trust in cloud based software, it's a whole different discussion than five years ago when I when I thought, Oh, I'll have this software for auditing, that's all cloud based. You could work remote that blew people's minds. Like, no, you have to be in an office. And now everyone realizes that, you know, offices are great connections face to face, just like we're having with this video chat here. And podcast is valuable, but it's not the only option for getting work done.
Amanda:
Stephanie Let me continue on that thread then. So for any other wannabe bootstrap ears that are listening to this podcast right now, what advice would you give to any entrepreneurial scientists that have a vision and want to start to bring it to life but are scared or have questions about it?
Lyn:
Well, the first thing I would say is if I can do it as a chemist and no financials, no finance background, then anybody could do it. I would say, though, clearly what I did to make me feel like I had a path forward, because part of it is just making sure that you feel like you're moving in the right direction because it can seem like you're running in quicksand, have clear goals of where you want to be next, smaller as well as large. And so therefore, you always feel like you're moving incrementally. Get a group of people that will be honest with you because there are times where you need to pivot or stop something that you think is going to move forward and it ends up not being something a path. Have those clear exit strategies as well, but really have the hard thought beforehand, but really of what you want and where you want to be. But also know that I was always told that investment and VC angel funding, all of that is the only way to go. It isn't. There are other options. It takes a lot of burden on yourself. You're using your own money, your own capital, your own credit score, but there are other ways to do it. So just really have faith in your vision and and critically look at it. But realize that some taking a chance. It's only a once in a lifetime opportunity. I'll never work for anybody else other than myself again. The love it.
Amanda:
That's great. I loved that. And I think, you know, given the current financial markets and things along that line, you know, the idea of having the power to do it yourself versus relying on VC firms is really inspiring. They're scary but.
Lyn:
Inspiring. And depending on your your area in the Cambridge area, especially the west, the coast, there are a lot of different local organizations that do look.
Amanda:
To.
Lyn:
Support women, own business, women and minority businesses, that there may be grants available available to you, there may be matching loans available. I have one of those that's a part grants and a part loan. I was lucky that I had the consulting firm that was a revenue generating business that fed the software, which is innovation side of the company, because even though there are two separate, they really are one entity sister companies, but realizing that there are perhaps areas of cash flow that are non traditional, meaning the non. Way really getting into your community and asking around about some of the available for cash flow as well.
Amanda:
That's a great advice and something that I would not have considered.
Lyn:
Yeah, going with investors is also a completely legitimate path forward. It's just you give up quite a bit of the company right off the bat and there's a you're expected to just exit really quickly to make the cash back. And I didn't realize that I was actually growing it more long term just based on how the the path forward was going. So for me, it was less of the quick turnaround as well.
Amanda:
So, Stephanie, I wanted to talk about something that you said, you know, I taking over the world, so I'm learning and you also mentioned being five years early. What are some of the challenges that you've seen in this space and what the predictions are for five years from now, ten years from now, and how fast the industry is just growing?
Lyn:
So what I say a lot is that we are a risk adverse people in this field. Is industry doing a very risky job. So those two things can conflict with each other. So when it comes to changing a large paradigm shift, which I foresee, I in all its different forms, because there's so many different types of AI, it's just not machine learning alone, there's so many different types. But I really feel that the next wave of what we're offering and what others are offering is human augmented intelligence, which means that the human's not taken out of it. It doesn't replace a job. It gives the exhaustive tasks to the computer through natural language processing and machine learning, other different types of AI. So that way we can do our job better and more efficiently. And because, as I'm sure you all are aware of, we're just getting more and more work coming down to us with less resources. So any tools we can pool. The difficulty is that cognitive bias, the fear that comes along with trying something new and also fatigue. Everyone's so tired. The idea of saying, I'm going to give you this tool, you have to learn it to use it. It will save you time, but it will cost you time up front. That's a hard sell to anybody that's overworked, anybody in general, but especially a populace that is overworked in a very stressful, life threatening. We deal with life threatening diseases all the time, right? That's where our data feeds into. So taking a pause on that to learn something new is difficult. So that's where I see. And then there's the regulatory bit, which is is not even that's not even right now the machine learning and it's not adaptive. So the ones that don't learn and change does have a path forward, but the ones that are adaptive and learn along the way and train themselves, there's actually really not a clear cut idea of how to have a path forward for that as a medical device.
Amanda:
So for our listeners who are pretty new to AI, how does AI help bio analysis like the work that we do, interpreting clinical data and things like that? Just taking a step back.
Lyn:
Okay. So for for what we built, there's all sorts of different ways that I could help with bio analysis or drug development. They have AI that helps look retrospectively at past data, big data and build target models and see which ones were successful and which ones failed to help. Maybe predict in the future where drug targets might help and even in modeling as well in some of the drug target and drugs.
Amanda:
So I can help even with early target discovery deciding what you want to do for your next therapeutic.
Lyn:
Yes, there are AI that are actually out there being developed that help early on all the way to. There is something I believe it's GSK has something that uses OCR and computer vision to track on labels where the consumer's eyes go to know what's an appropriate label for consumer. Yeah, it's so it goes from soup to nuts to what is on the shelf to when it's being discovered in the lab where there are various different AI out there being developed. So what ours does though is there's the most basic AI is called an expert system. It's a yes no if then statements, it's a mapping out of a thought process which was developed in the fifties. So it's been around for a long time. And so the idea here is we can map out over and over again how data may or may not pass or fail the different criteria that we have. We get exhausted as reviewers when you're giving it in aggregate, right? Because you might lose track of how table one impacted table 100 because we're here. A computer does not get tired. It just doesn't. And it can do it in milliseconds. So the idea is to build that expert system over years and multiple consultants, multiple biomedical chemists to have basically such a force multiplier of multiple reviewers reviewing your data at once. Then the hardest part was building it. So that way if I have to go into a system and input by hand, all these different things like concentrations and I'm not going to want to do it. So we actually the hardest part was getting it. So it was platform and form agnostic, so you could just load in like three different things.
Lyn:
You have to have your your name and where you're going and what you want to do. I jokingly say, So you have to have the compound name, the project name, and what type is it? Small molecule validation, large molecule. That's all you need to input everything else. It's smart enough to go to the tables and the document and pull it directly from it and do everything itself. That was the hardest part and that's where the machine learning came in and other things because it has to be smart enough to know where to find it and where to pull the data based on just giving it example after example. So the idea here is to give you an auditing tool so that way you can spend instead of 7 hours of your eight hour day finding the problem and one hour solving it, if there are any, it's the opposite or freeing up your time, if there are no issues, and then finding it during development or before the end of the validation or sample analysis so you can go to your clients and troubleshoot together or not be caught unawares if anything, or repeat something that you realize needs to be repeated before it gets out of the production side of things. So that's generally speaking, how we've built. And then there's a cut point analysis which we've automated as well. Anything manual we try to automate. And so to provide it as well to the biomedical chemists so they can understand the stats and have some power, but lock it down as well, meaning it's just no systematic remove removal of anything. It's statistically driven analysis.
Amanda:
So that way you would get away from the issues of my favorite way of calculating cup points versus Amanda's favourite way of calculating Cup points. And if we're both doing Cup points for different assays but are for the same drug, it can lead to inconsistencies.
Lyn:
Exactly. And it just makes it, as I said, it happens in seconds as well. So sometimes there's that you throw it over the fence to your client or to your bias. That's and it comes back. And so yeah, there's the consistency. It's also along with guidance and you have a PhD statistician on hand to help this build it and they're there to advise as well. So yeah, the consistency, the quickness, I mean it just is all there at your fingertips on the cloud as well.
Amanda:
Super impressive.
Amanda:
Well, thank you. I kind of wanted to talk about the application and biomarkers and biomarkers are such a big part of drug development and the amount of data that's being generated, especially with all these very large multiplex platforms doing that early discovery and being able to find a good biomarker to carry through with with the drug application. How how does that also play a part with AI?
Lyn:
So there's actually I had a conversation yesterday with someone about the idea there's there's the data review process, which we have the toggling of the different acceptance criteria as well as for the multi different analytes as well. But there is the idea of taking all of the data and tracking it through a AI with accordance of the actual disease state with the patients. So there is a lot of that work going on. The idea of having more like real life, real time analysis because a lot of the times you do your biomarker analysis or even your inclusion exclusion criteria based on the status of the baseline of your patients. But actually cancer changes, it constantly evolves. And if you don't check every once in a while, you know, periodically through the treatment of that patient, their actual biomarkers, their targets might have shifted. And so your your actual treatment should shift. And we're not doing that at all. We're not monitoring in real life, real time the actual progression of the tumor morphology and therefore the biomarkers that come after it. So we have the informatics, we have the ability to consume the data and to to also track it and provide insight and trending. And so that's where I see that coming. It'll help the treatment of the patients in more of a real time. It's not like 20 years ago where we didn't even know how we map a genome because how are we going to handle that much data? We can handle that much data now. And so that's where I see AI coming in, not only in oncology, but cell and gene therapy and understanding these data and. Being able to in more of a real time treatment. And the tests are no longer thousands of dollars. They're hundreds of dollars. And so hopefully that costs will go down as well. So the idea there is to fully track more in real time.
Amanda:
I mean, I love the idea of bringing down drug development costs because that's one of the big issues, right? Like when we talk about these very sophisticated drug modalities, it's amazing what a cell therapy or a gene therapy can accomplish for an unmet need. But at the same time, it costs so much, it takes so much time to make these therapies. And then that pulls out on the back end is being patient burden. So the idea of using AI and some cian learning and automation to help cut down on the time it takes to get the results is really powerful. And I think ultimately we'll have big ramifications for drug and pharma industry at large.
Lyn:
And I think we're going to be shifting to where majority of oncology patients right there enrolled almost as if a rare disease in themselves because their cancer is different than it's no longer all lung cancer or solid tumors are all together. They're they're stratified by by different criteria. And so you'll have a triple negative breast cancer patient in the same as another non-Hodgkin's lymphoma. Who knows? But. So that stratification is so different now based on biomarkers and the targets of the drugs that we have been developing. So I'm hoping that by having some real life feedback and following the patients and modeling that both and you can also retrospectively look back, there's a lot of data that we can pull from to see where it was successful and where it wasn't. You know, there's HIPA compliance, There's there's all sorts of different concerns around safety of personal data, but there are large amounts of data that we can mine as long as we come together to have more full transparency about treatments and drug platforms and things that were successful and not successful. I think that's where we need to get compliance together to really kind of take our egos off the shelf and and talk together about the patients and treatments that worked and didn't work because most of the time I've learned most from my mistakes, not from my successes. So that's where that data would be extremely critical as well to meet unmet medical need, get it to the patient faster, get go, no go decisions quicker because that's where you save a lot of money.
Lyn:
If you look back at the 2014 Tufts paper on costs of how how much it cost to develop a drug, it says I think it was $2.67 billion to develop one drug. Now, there's a lot of debate about whether or not that's a true number or not. And it was an aggregate of across different programs and different types. But the idea there, though, that was salient to me is that 40% of the cost is opportunity cost. You're losing opportunity of working on something that would work by by working on something that doesn't. So 40% of that $2.67 Billion was just actually because you were working on the program. That didn't wasn't successful. So if we could recuperate a lot of that cost and make the decisions earlier, it's just getting people to be OC to be told when they're wrong or right. It's not an easy message like our software is telling you, auditing your data. People don't like their data audited. Right? And so how do I make them okay with with this like it's to help you not to, to lecture you. That's a lot of it's cognitive bias and getting over that fear of adoption. There's a study where they list all the different things of reasons why I wasn't adopted. It was only 1% because it didn't work. All the other reasons had to do with training and staff and knowledge. 1% was because it didn't work.
Amanda:
I totally believe that, Stephanie, because I know every time that there is a software update for my phone that is rolled out, I am angry and filled with rage that I need to learn the new system. So I know.
Lyn:
I know it's it's old. I know. And so it's one of those we could sit there and as a community, try to change how people are thinking. And I think we try to do that by education and and awareness and just becoming comfortable with sort of different softwares. But also there's a way of making it more consumable as well. Right? How do we make the software more easy to consume by the end user? And part of that is if you have a whole bunch of software developers creating creating widget, but then you don't have the end users, they're saying, I wouldn't use it, then that's not going to be helpful. And so having those communications are key as well, because that's that's tough love that you need to take in, even if it's the greatest thing. If no one uses it, it's not helpful.
Amanda:
Right.
Lyn:
So.
Amanda:
You and I totally agree. I think evolving our thinking, it just has to happen because even ten years ago I feel like personalized medicine really wasn't. It was it was a little out of reach. And we're here. The data is here. The ability to mine that data is here and to apply it and drug development is just kind of blows my mind just from ten years ago to where we're at now.
Lyn:
I mean, it hit even home closer to my family because my my dad's past. But he had stage four lung cancer. And originally they just would treat him with a cisplatin and they put him in a clinical trial for that, for a monoclonal, which was normally not how you treated it. And he ended up living for 12 years. They gave him one year and it's all because of just looking at his markers, putting him in actual trial that normally he wouldn't have been slotted in as a lung cancer patient. But because of the morphology of the tumor, it was a good fit and he lived for 12 years. It was amazing. It was extremely aggressive cancer that was knocked down by the treatments. So that change and think, you know, 15 years ago, being on the cutting edge ten years ago was completely changed the way we're even enrolling more traditional oncology trials. It's crazy.
Amanda:
And if I can help us to better accumulate patient data to make that decision, to identify that patient that might benefit from that therapy, you know, it's going to make a huge difference.
Lyn:
It will. It? Well, the issue is sometimes, even though we do have a lot of data, there's big data perhaps, but there might not be. So big data is if you're trying for machine learning unsupervised, trying to teach a system what's a duck look like. So you just feed it a whole bunch of pictures of ducks and dogs and it categorizes them and then it learns itself. But that's because you have millions of pictures of a duck. You might not have that volume. We have a lot of data, but it might not be big data. So but there might be some big data, sort of data repositories we can pull from as well. It really depends. But there's a whole bunch of different approaches that will need to take that we can do both supervised and unsupervised learning and pull the data in aggregate and look retrospectively and then hopefully project that forward and really learn, learn from past data sets. So yeah, so I think that's where we struggle is that we have a lot of data, but it's not tons so that we can't teach the system what a cat is on its own, we might have to supervise it.
Lyn:
There's also the fact that when you start it, when it starts to learn on its own, you can't always break it out so that, you know, how did it learn how to do that? So sometimes it just happens and it's hard to then translate it to human action. How can we as humans. So that can be a little scary at times. You know how the computer gets better at things. We don't really know how it does, but the best relationship that we've seen is they even did it with chess. They showed like an AI against a top chess player, and then you have the person against another person champion. The best combination was the champion chess player With AI together, competing those two together was better because the the strategy and the fact that a software is going to know like a bottle standing up, it's going to know it's sitting on its side. It doesn't understand the context of everything is going to spill out because you laid that on its side. That's where humans need to come in. That impact.
Amanda:
Kind of reminds me of the Terminator.
Lyn:
A little bit of it, right? Yeah. It's just the it's not going to understand the reads of a poker player, Right? It's not going to understand some of that because it's not going to have that human intelligence behind it as well. So the two working together usually works out the best.
Amanda:
So, Stephanie, if you weren't doing science, what would you be doing?
Amanda:
I'd be a chef.
Lyn:
Oh, most chemists are probably good cooks. Not always, but usually we're a good cook or a baker. And I love. Love to cook. So what is your signature dish? I make a killer enchilada because I am a Southern cowgirl. Yeah. Flank steak and braised all day is really good. So I do. I do love that. But I also make really good crepes in the morning. I like big breakfasts. So in the weekends, when especially when I'm traveling a lot, I'll cook like a big breakfast for the family and do like a French toast or crepes or something like that. And so big breakfast is also a big thing of mine.
Amanda:
It's funny that you mentioned that you like to cook because one of the questions that I had for you is, as the founder of this company, starting to make headway into AI and bio analysis, that must take up a lot of your time. So how do you maintain balancing your life between the personal and the professional? Any tips or tricks? I always try to glean what I can from people.
Lyn:
So it is very hard. I mean, I'm not going to sugarcoat it that I mean, people say you can have it all and you. You can't. You have to choose, but you have to set up boundaries. And so there are I'm not always very good at saying no, because I do love what I do. I do and I and I love problem solving and it can get like a rather my husband jokingly sometimes says I'm like, the Benedict Cumberbatch in the shakes is for where my mind goes crazy. He's like, You're off somewhere having a having a moment when a data set comes through. So just making sure to like I practice yoga. I used to run long distance. I don't do that as much anymore, but there's things that are very meditative to me that I try to make sure to incorporate even on the road when I can. And then just having a really strong support system. That's part of the reason why we moved back to Kansas is because we were all on our own on the East Coast. And so this was really helpful to at least know that my husband and my kids had family nearby if anything was needed. Yeah, And also knowing that. There's a lot of benefits to having your own company because you are your own boss. You are you make your own decisions. But there's a lot of fear in that. And. But there's nothing wrong with also not liking it. So if you go into it and you end up not liking it and just self awareness of I like it when I'm in a group that has more the structure and I'm not the the buck stops there person, that's completely cool too. So it's also a little self awareness because there are a few times where I almost said this is too much because it is a lot of pressure, so.
Amanda:
Understood. Thank you.
Amanda:
Thank you so much, Stephanie. I think it's been a really awesome conversation. We really appreciate you sharing with us about science, sharing about what it is to take risks and how that evolved into success. Also, appreciate you kind of giving us a beat on where AI machine learning falls in the industry right now and your thoughts for the future and where we're headed. So thank you so much for taking the time.
Lyn:
You're welcome.
Amanda:
So that is all for today's episode. If you enjoyed listening in. Be sure to subscribe on Apple Podcasts, Spotify, or your favorite podcast app so you never miss a conversation. If any of our listeners or other scientists in the industry, you'd like to talk more about how we can help you with your drug programs, please visit our website at bioanalytical and use our Speak to a Scientist tool to get in touch. Or if you'd like to hang out with us more casually outside of the podcast, we have many webinars and other presentations available for your enjoyment and education. Just visit Biogen politics.com to see what's coming up and how you can stay in touch.
Amanda:
Thanks for listening to the Molecular Moments podcast.
Sonix has many features that you'd love including enterprise-grade admin tools, collaboration tools, automated subtitles, share transcripts, and easily transcribe your Zoom meetings. Try Sonix for free today.
Share This 
