image: thumbnail.jpg
Liyang Diao @ ROME Therapeutics | Data Science Hangout
Speaker bio: Liyang Diao is the Director of Data Science at ROME Therapeutics, a biotech focused on harnessing the swath of the human genome comprised of repetitive elements, which to date has been overlooked in human health and disease. Her industry experience also includes work in the microbiome field at Seres Therapeutics, where she was part of the team that brought the first oral microbiome therapeutic through FDA approval (VOWSTTM), and where she found herself particularly drawn to emerging fields where genomic data play a critical translational role.
________________________
āŗ Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here:
Website: https://www.posit.co
LinkedIn: https://www.linkedin.com/company/posit-software
To join future data science hangouts, add to your calendar here: https://pos.it/dsh
We'd love to have you join us in the conversation live!
Thanks for hanging out with us! š
Transcript#
This transcript was generated automatically and may contain errors.
Hey everybody, welcome to the Data Science Hangout. If we haven't met yet, I'm Rachel, I lead Customer Marketing at Posit. I'm excited to have you joining us today.
The Hangout is our open space to hear what's going on in the world of data across all different industries, chat about data science leadership, and connect with others facing similar things as you. And we get together here every Thursday, every single Thursday, unless it's a holiday, at the same time, same place. So if you're watching this as a recording and you want to join us in the future live, there'll be details to add it to your calendar below. Just make sure it adds it for 12 Eastern Time, 12 to 1 Eastern Time, so you can join live.
If this is anybody's first Data Science Hangout, I'd love to see you say hi in the chat so we can all welcome you in and say hello as well. We're all dedicated to keeping this a friendly and welcoming space for everyone and love hearing from you no matter your years of experience, the languages that you use, titles, industry. It's also totally okay for you if you just want to listen in here and be part of the party that's happening in the Zoom chat. But if you want to jump in and ask your own questions or provide your own perspective on certain topics, you can raise your hand on Zoom and I'll call on you to jump in.
You can put questions in the Zoom chat and feel free to put a little star next to it if you want me to read it or I can call on you to introduce yourself and add some context. We also have a Slido link, which I'm sure Curtis or KJ will share here in the chat, where you can ask questions anonymously too.
I love getting to use this time with everybody here to share a few other notes of upcoming events too, so just real quick. In the Hangouts, we've had a number of questions on best practices for management and stewardship of data. So I wanted to let you all know that we are having a special event in May on this topic with two previous featured leaders, so Jamie at Plymouth Rock Assurance and Dan at Biogen. And so I'll share this event registration link in the chat here if you want to check that out. That's going to be on May 15th, so I just wanted to give some people some advance notice.
And then I also host a monthly data science workflow event, and it's the last Wednesday of every month. And so this month, Julia Silgay from our team is going to be joining us to share how to develop and deploy a machine learning model with Posit Team. And so if you want to check out those monthly workflows, I'll just put the add to calendar link there too. Cool.
Veronica, I know you had something you wanted to share with people before we get started as well. I want to jump in. Hi, everyone. I just want to introduce myself. My name is Veronica, Veronica Clark, and I'm actually new to Posit. This is my one-month anniversary. But the great news I want to share, I'm actually a principal UX designer, and I want to leave it open if anybody ever wants to reach out and show me your workflow and how you're using Posit and areas of improvement, I am here for all of that. And it's very easy to reach out to me, it's just Veronica at Posit.co. So yeah, hopefully we'll run into each other as I continue learning and working with these products and improving them for y'all to use. So just wanted to introduce myself. Thank you.
Well, thank you all again for joining us today. I am so excited to be joined by my co-host for the day, Leanne Dao, Director of Data Science at ROME Therapeutics. And Leanne, I'd love to have you kick us off by just introducing yourself and sharing a little bit about your role, but also something you like to do outside of work too.
Introducing Leanne and ROME Therapeutics
Sure. So my name is Leanne, I'm at ROME Therapeutics, and I started here in my current role about seven or eight months ago, so I'm still relatively new. So ROME Therapeutics is a small biotech company based in Boston. And what we are studying is what we call, well, what some people call the dark genome. So if you think about the human genome, we're all pretty familiar now with the genes that code proteins and have functions that are relatively well studied. But there has been, I think, an increasing appreciation for things that are not protein coding genes. And I think that this is an area of study which is, it's challenging from like a computational perspective. You know, how do you quantify these kinds of elements which have special challenges which we can talk about? And then on top of that, how do you, you know, relate that to potential new therapies? And so I think it's a really exciting space. I'm really excited to be here.
My background, it's going all the way back to college, I guess. I studied mathematics, and I think I minored in Chinese language. So I spent some time in China, you know, just, I think I spent an entire semester just taking language and literature classes, which was really fun and totally different. I think that was my final semester in my senior year. So that was very exciting. But before that, it was really just math all the time. And the idea that I had had when I was in college was to just pursue math. And I was really not interested in biology, to the chagrin of my mother who wanted me to become a doctor and study pre-med and go that track, you know, all the way to the end.
I did try for, I think, a semester or a year, and I thought that biology was just, it's just a hot mess. No one knew what was going on. And it was just very, it was not very satisfying compared to the kinds of problems that I was trying to solve in my math classes, which were clean and well-defined, and you can, you know, write a proof and it was either true or it wasn't true, right? In biology, you don't have those kinds of, it's not very binary. And so I, you know, I really was very set on this all the way through to basically my senior year. And then in my senior year, I had to think about what to do next. And I wasn't ready to get a job. So I thought, well, I have to go to grad school then because I have no other skills.
So I was applying to graduate school and I applied to a bunch of math programs, but then I also ended up applying to some bioinformatics programs, even though that wasn't something I had really studied before. It was something that I sort of became interested in, in some of these summer undergraduate research programs. So I was a math major, but then I tried, you know, for a month or two to study a little bit of genomics. And with genomics, I thought that was something that was cleaner than biology, but was still applicable in the real world. And so that was, you know, it kind of piqued my interest and it felt like it was the best of both worlds because towards the end of my undergraduate career, I was beginning to think that pure mathematics, which is what I was studying, it wasn't applied math, it was pure math, was becoming less satisfying because I just couldn't see how what I was studying day-to-day had any impact on, you know, what I saw day-to-day, you know, the lives that we were living. So that became less interesting to me.
But I applied to a couple of bioinformatics programs and, you know, I got into a few of them and I decided just to make the jump. You know, I had no idea really what I was getting myself into, but I was ready to pull the triggers, so I just went. And that's really how I ended up in computational biology slash bioinformatics. So it's kind of a long-winded story of how I got here, but I'm really happy to be here.
I think it really makes use of, you know, some of my skill set from my math background, but also, you know, I've come to appreciate more and more that even though biology, I guess it's more that I appreciate the complexity of biology now more than I did before. I think before I was really turned off by it and kind of scared and I didn't know how to interpret anything. But I think over the years, I think that unknowingness has become more exciting rather than scary.
But I think over the years, I think that unknowingness has become more exciting rather than scary.
Thank you so much for that intro into your journey. What about something that you like to do outside of work too? Yeah, that has really changed a lot over the years. You know, now I have two little kids. They're both in daycare, so they're not school-aged yet. And I spend most of my day sitting in front of a computer, you know, doing this. So what I like to do outside of work is really to get outside, to get moving. So I run. I've been running for maybe like 20 plus years now. It's really, I'm not fast. I don't run to race. I run to be outside. I run for the feeling of running.
You know, I think one of my best graduate school friends was also a big runner. And I think that she put it really well when she described why she enjoyed running. You know, she really treated it almost as like a meditation activity. We used to go for these long runs and you just get all of the stuff that you were thinking about that was percolating in your head, you know, all day or all night, you get it out and you get the endorphins instead. And it really helps to clear my mind, especially when I'm like really stuck on a problem and I'm feeling really frustrated. It really helps.
Yeah. It's fun to see what are the common hobbies among this group, because running is definitely one of them. And then playing music is definitely. Yeah. Yeah. I've been playing piano for many, many years, but I started drifting away from piano when I started doing more computer work because after the end of a long day, I don't want to sit again in front of something I want to be moving around.
The dark genome explained
Well, I was wondering, you mentioned a bit about the dark genome, but for some of us without a biology background, would you be able to explain a little bit more about the dark genome and about the the work you're doing at ROME? Mm hmm. Yeah. So I think probably everyone has learned in school this central dogma of, you know, we have our DNA. It encodes, you know, genes which get transcribed into RNA and then RNA gets turned into protein and that the protein is the stuff that does stuff in our body. And I think that that has been the dogma for many, many years. But the the percentage of our genome that encodes these protein coding genes is actually really small. So most of our DNA is not coding for that. And I think that we're, you know, we're still sort of learning what else is there.
So these non protein coding genes, I think, for a long time have been described as junk DNA. But we called it junk because we didn't know what it was doing. And so at ROME, you know, we're very specifically focused on this class of elements called repetitive elements. So, you know, we have like one copy of gene A, maybe two copies of gene B. But then there are these elements where we have like hundreds of thousands of copies of basically the same element. It's like a gene. It may or may not encode protein. And some of these elements have the ability to lift themselves up and make new copies in our genome. It's very cool. But we don't really understand like the full implications of what they're doing.
And so we think, you know, especially in the last, I don't know, maybe five or 10, 10 years, people have been starting to appreciate that they're playing a role in potentially cancer, in autoimmune disease, in inflammation. And so at ROME, we're, you know, thinking about these elements and not only what roles they play, but, you know, in which human diseases could they be causal, or maybe a biomarker of disease, and where we can potentially impact human health for the better. Is that helpful?
Yeah, that is helpful. I know, Bill, you had, I think, a follow-up question here on dark genomics. Bill, do you want to jump in? You want me to read it? I can read it. But the question was, with dark genomics, are you using the same widely used suppliers of sequencing machines, or are you developing new software related to new hardware? Yeah, so the sequencing technology itself, we're primarily using the same sequencing technology. So, you know, there's short-read sequencing and long-read sequencing, there are kinds of pros and cons of both for looking at these repetitive elements. I think the challenge is, I mean, in part the technology, but really what we are focused
on is downstream of that in the quantification algorithms, and the algorithms to detect where new elements are being inserted, which weren't there previously. So a lot of this is algorithmic development. And you know, I didn't build any of these algorithms, these were here when I joined. But we have some really, really excellent computational scientists who have been at Rome for, you know, a couple of years now, who have really built up this platform.
Yeah, Travis, I see you just put a question in the chat, want to jump in next? Sure. Hey, it's cool to meet you. I spent some time thinking about long non-coding RNAs in prostate cancer in a very academic setting. And since then, I've moved over to kind of CRO world, much like where you operate at Rome. I'm curious about your business pipeline, like the research pipeline. So what proportion of it there happens by virtue of kind of computational discovery only? And then where does that data come from? Like, so in the in academic setting, it was a lot of kind of epidemiological studies and specimens collected, we do deep sequencing, and then we kind of try to infer through ML and other techniques, what kinds of targets in the dark genome might be relevant. But you'll have a different kind of pipeline, you do a lot of my studies, is there other in kind of in vitro work that you do?
Yeah, so it depends on, I guess, how far along the asset is, you know, so we have, you know, assets that are approaching the clinic very soon. And then we have assets that are like early, early development, right. And so depending on where we are, in terms of how far along clinically, I think it changes, you know, what kind of research we do. So for our lead program, our PT 2950, we're going after autoimmune disease, and specifically cutaneous lupus. And so, you know, a lot of research has already been done there. And we've done like, for example, a lot of in vivo and in vitro studies, and some ex vivo studies, which are really interesting, where you actually, so this is they're like a CRO that does this, where they take these skin samples, which have been excised from patients undergoing like cosmetic surgery. And then they can, you know, grow, not necessarily grow, but like keep them alive in a in a dish. And you can treat them with different things with like, with UV or your compounds or whatever. And then you can sort of see what changes in terms of their gene expression, you can even profile, like with imaging technology, like how the skin changes over that period of time, it's really cool.
For like early discovery things, when we're looking for new indications and new targets, that's where I think you have to do, you have to kind of take a step back, because you don't know exactly what you're looking for there, right. And so I think that in part is driven by computational discovery. So we've processed, I don't know how many, like 10s of 1000s of samples, maybe 100,000 samples at this point, of public data, all across different kinds of indications, all across your, you know, your your gold standard data sets, like TCGA, GTEx, and you know, those kinds of data sets, and a bunch of different indications from studies in public databases like GEO. And then there, we look for, you know, any indication of activity of the targets that we're interested in, and also genes which are related to activity.
That's cool. So in your team, do you have like a small army of people who are just cranking on TCGA, GTEx, and GEO, like all day long, sounds like the academics dream, you know, We need a bigger, we need a bigger army. Army's never big enough. But yes, more or less. You know, that's, I think that's one of the fun slash frustrating parts of being in a small company. Like Rome is small. I think in total, now we have like 30, 30 something people, and the computational group is me and three scientists. And so you can imagine the amount of work that it takes to develop and then maintain the pipeline, and then do the analysis, and then like continually trying to improve that pipeline. It's a lot of work, and we get stretched pretty thin. So prioritization is, is really important. And your priorities are going to change from day to day.
Cool. Thanks. And not to jeopardize all the quick, but then the skin, the skin, what xenographs, I guess you call them maybe, like, what do you do? Do you like give them sunburns and stuff? Yeah. I was, I was at a colleague who had like a whole basement full of, I mean, it's not a basement. It was like a research lab full of many brains. He called them. It was very X files. Yeah. And you always wonder, like, are there, do they have consciousness? They're like living in these little petri dishes brains. It's like weird. Yeah, that's very cool. Now, I had not heard about this kind of assay until it was run and we were analyzing the results just maybe a couple months ago.
Prioritizing projects at a small company
I know you, you actually just talked a bit about prioritizing and that was actually a question I wanted to ask you here. So out of a company with 30 something people, how do you prioritize projects and allocate resources? It's a really good question. And the answer should not be haphazardly. You know, I think that you really need to have a high level understanding of what the timelines are for the company and what's driving those timelines. Right. So you have to kind of take a top-down approach and okay, I'm going to, I'm going to diverge a little bit and tell an anecdote.
Both of my parents are university professors. And so my dad teaches mathematics, so pure math. And my mom is a professor of statistics. And I remember when I was describing, you know, my day-to-day because they've only been in the academic world and they're like, you know, what do you, what do you do every day? And I'm like, well, I sit in like five or six hours of meetings every day. And they're like, what do you, why do you need to sit in so many meetings? Like, what are you doing? And I think a lot of it is just, you know, getting everyone and all the teams, like these very disparate teams doing very disparate activities on the same page, because you have to drive forward these programs where you need biology to run biology experiments. You need computational folks like myself to run computational experiments. You need chemistry to make the chemical matter. And trying to get everyone on the same page to agree on what is the best path forward.
What is, you know, what are the go, no-go decisions and what are the timelines in terms of, yeah, so what are the timelines and how long can your company survive for, you know, especially as a small company? That part is really important. So that takes a lot of effort. You don't necessarily have to think about that in an academic setting as much. So even though my dad was the chair of the math department for many years, he had to deal with, you know, some of this, what they call politicking, right? Getting everyone on the same page. But it's like Professor A and Professor B are working on two completely different things. They don't have to, you know, get together and talk about what are you doing today? What are you doing today? What are you doing next week? They don't have to align on those kinds of things. And it turns out that alignment takes a lot of time.
And it turns out that alignment takes a lot of time.
Transitioning into data science and biology
Absolutely. Nellie, I see you had a question in the chat here. Want to jump in? Sure. Thanks, Leanne, for sharing about your journey. I studied biology in undergrad, as well as economics. And now I've been working in finance in a data science role for a few years, about five years. And something I've always thought about was transitioning back over to the intersection of data science and biology. But I do know there's definitely a lot of need for higher degrees. And I was curious what you thought, you know, are certain things that individuals should take into account if they're interested in moving over to the sciences and away from a traditional data science role? Traditional finance, I guess.
Yeah, I don't know much about finance. But I do know that people, at least early on, you know, when I was in grad school, people made this change between going the finance route and going the bioinformatics route. They made it pretty fluidly at the time. I think that that's probably different after you've been in the workforce for a little while. So I think after that, it becomes a little bit more challenging, but it's not impossible. Especially if you're willing to start in like an entry level role. I think that's entirely possible.
And I think that, and I'm biased here, but, you know, one of the things that I love about working in startups is that, you know, you don't have to have a lot of experience. One of the things that I love about working in startups is that, well, okay, I don't want to say the wrong thing, but I think that one of the benefits of working in a startup is that you can start a role and not have like very, very deep and narrow knowledge in a particular role. I think what you typically need to be able to do in a startup environment is to play many different roles and start up very quickly.
So I'll give a little bit of an example from my own experience, which is not quite what you're, you know, what you're talking about, but I think it has some relation. So in my first industry job, I was at a place called Ceres Therapeutics and we were very, I mean, we were a microbiome therapeutics company. And so I worked there for almost six years. And when I was there, everything was about microbial profiling. And, you know, we were like processing stool samples. We were thinking about what is the best algorithm? What is the best database for determining what kinds of bacteria are in a sample? Nothing like human data related at all, basically. And towards the end of that, I was thinking, you know, I wanted to broaden my experience a bit to, you know, working in like to learn more about oncology, to broaden outside of the microbiome space. And, you know, it's not like I could jump directly into, you know, like a principal scientist or senior director of oncology at one of these small startup companies, right? Because I didn't have the experience. And so I ended up going to a small company that didn't have a bioinformatics function at all. And so I got to, you know, A, learn how to build up that bioinformatics infrastructure, spin it up in the cloud. But B, at the same time, I had a mentor who basically took a chance on me and gave me a little bit of time to sort of ramp up on oncology. You know, how do people think about, you know, problems in oncology? Specifically, I was in this space called the Antibody Drug Conjugates. So that was completely out of my wheelhouse at the time.
But it's not something that you can't learn, right? So I think, you know, what's really important is to understand your own capacity for learning new things. And I think that if you're confident that you have that capacity, that you can do it. It's just how much effort is it going to take? And are you willing to commit that effort? But I'm a strong believer in that, especially in startup spaces, because, you know, one day this company is going to be here, and one day it's not, you're going to have to go somewhere else and probably pick up new skills while you're at it. I know that's probably not like super helpful, but I think it's possible is my answer. And it's just about finding, you know, the time to do some training on your own, but also finding the right fit in terms of a new job, you know, in the space that you're interested in going into.
So I think, you know, what's really important is to understand your own capacity for learning new things. And I think that if you're confident that you have that capacity, that you can do it.
That's great. Thank you so much. And I think also once you know yourself that you have that capacity to learn, how do you then show the hiring manager at this other role that you do have, and that's the benefit? Yeah. A lot of it is trying to draw parallels. So maybe you don't have a specific skill. You haven't done this particular analysis before. But in data science, you know, one of the great things is that if you're working with data, you just have to know what kind of data it is in order to know how to treat it and how to do the analysis, right? So maybe I haven't, you know, done an analysis with human gene expression before, but I've done a lot of bacterial, you know, composition analysis. So you can draw those kinds of parallels. And I think it's easier to do that in the data science space than in some other spaces.
Thank you. Svetlana, I see you had a question a little bit earlier. Want to jump in here next? Yeah, thank you. Yeah, so my question was about computational biology and bioinformatics versus data science. So my background is actually a bit similar to yours. I had a computational biology slash bioinformatics degree and then a little bit of postdoc experience. And then I transitioned to data science. But I feel like that it's, for me, it's always kind of, so I'm not doing bioinformatics anymore because I feel like I probably don't want to, or maybe I was a bit too traumatized with like the academia experience and like working in the lab and like all of this. But I sometimes think of going back.
But I'm curious because I was also like looking at your profile, right? So right now you're director of data science, right? But so do you feel that this role is different or is it because like you have data science in your title or it is still more like computational biology but just named as a data science?
Definitely the latter. This is a trend in the industry that I really only noticed in the last couple of years. You know, five, six, seven years ago when I was getting started, everything was comp bio, bioinformatics. And then in the last couple of years, I started seeing everyone's title started shifting towards being data science.
I don't know what was driving that trend. I still consider what I do computational biology. You know, maybe people want to say data science because it seems to have more like broader implications. So we do more than just, you know, genomic pipelines, for example, like we analyze all kinds of data. So in that sense, OK, you know, data science, that's fine. But it's definitely still the same thing.
Yeah, interesting. But so would you say that so at your company, for example, yeah, there is no other data science, right? No, there isn't. The kind of separation that I've seen, which is more typical, I think, is within is like between computational biology and then bioinformatics slash bioinformatics engineering.
So, you know, for so this was the case that when I was at Ceres, we had a computational biology group, which is more focused on translational analysis. And then there's the bioinformatics group, which is more focused on building the infrastructure and the algorithms. So I started out in the bioinformatics department there and sort of slowly transitioned over to the computational biology group. That is more of the distinction that I've seen in the past.
Keeping up with the field
But it was how do you keep up with the technology and knowledge in the fast developing area? Yeah, that's it. I mean, that's a good question. So if so, first of all, you know, in a lot of companies, we have resources that go out and find patents and IP and interesting what appears to be, you know, interesting related publications every week or every month. And we get that kind of report.
But also what has been really helpful, especially when you're in like a new field, is to just set up a ton of Google Scholar alerts. So I have Google Scholar alerts for all sorts of things. And then every, you know, it should be every day. So it's not every day. So it does pile up over time. I think I have like 90 unread ones right now. But I try, right? I try to get through the pile every couple of weeks or so.
And so you can that way you can really keep on top of like cutting edge, what is being published? What are people saying? It's like the biology side, the clinical side and the technology side. I think that that is actually really helpful and low lift, except for when you're, you know, trying to go through your inbox.
Yeah, I think I think that actually ties into something you and I had chatted about a little bit before, but you said in order to be the best data scientist, you can be you have to know the right questions to ask in that space. And other ways that you kind of went about that when you first started in your role.
Asking the right questions
Yeah, I think that that is an area of growth that most people go through, you know, as they become more senior in their roles. So, you know, when you are fresh out of grad school, I think most of the people who, you know, when you first graduate, you have a lot of really sharp technical skills. And so you really just need someone to tell you, you know, solve this problem. Tell me, you know, very basic question. What are the differentially expressed genes? What are the pathways that are differentially expressed? And everyone can do that, right?
If you're coming from a bioinformatics background, I think as you progress and you start, you know, or in my case, moving away from just the bioinformatics and going into computational biology, into the translational questions. At some point, someone like there isn't going to be someone to tell you what are the questions to ask, right? And at that point, you either, you know, need to find someone who can do that for you, or you need to start thinking of those questions yourself.
And at that point, you either, you know, need to find someone who can do that for you, or you need to start thinking of those questions yourself.
I think that part can be pretty challenging because you have to understand not just what your tool set is able to answer, but what are the data that are available? You know, what are the relevant biological questions in a particular indication or in a particular space that's relevant for your company to help drive the science forward? And that, I think that takes, you know, it can take a while to learn to pick up.
But I think that's where people who have the biology background, you know, people who come into bioinformatics have all different kinds of backgrounds, right? So, some people come in from a math background like myself. Some people come in from a physics or a chemistry background. And some people, I've known actually several people who come from like an ecology or a biology background. And so they actually, I think, have the benefit of having already been trying to ask those kinds of specific questions, which, you know, for someone like myself, who really avoided a lot of biology classes growing up, you know, I had to sort of learn that on the fly.
Data science beyond biology at small companies
Kevin, I see you have a question in the chat. Want to ask that one next? Yeah, sure. Yeah, my question is kind of related to what you were talking about a few minutes ago about being the data science director or, you know, having a title of data scientist when you're really doing computational biology work. But yeah, I was just curious, you know, since it is a small company, do you end up getting pulled into or seeing opportunities for efficiencies in terms of like business informatics?
Or, you know, I find as someone, you know, who's doing more scientific work, but having a data science background around a bunch of people who aren't data scientists, a lot of times I'll be interested in, oh, I could solve this administrative problem that people are facing, you know? And sometimes that I think is a good idea. I'm just curious, yeah, if you end up getting pulled in that direction, how you feel about that, if that's something you enjoy or something you saw as a distraction.
I will say that I probably have not had that many opportunities to do that because we've had so much activity going on in the data science group that we just haven't had the bandwidth to look into that.
You know, there are some examples where, but it's all like related to data science, like in terms of, you know, cleaning up our data science workspace, like our Google Cloud workspace, and working with a chemistry group, and trying to, you know, streamline things in terms of where we run our computes and how do we spend this money. That's something which is kind of an administrative task that we do get pulled into, but that's, again, still data science related, right?
I guess the one example that happened kind of recently is, you know, sometimes we get, we are asked to summarize a bunch of, like, conference proceedings. And so, you know, I had at one point at my previous company, I was the only data person there, and they had sent a couple of people to different conferences, and they had brought back all of these proceedings or presentations in PDF format, and they were like, well, we want to summarize all of these things. So, there were a lot of documents and, like, hundreds of conference proceedings. And, you know, a friend of mine who is a scientist was complaining to me. She was like, well, I have to read through all of these presentations and provide a summary. And I was like, well, I could probably help you a bit with that. And so, you know, we can pull the text out of the PDF and then summarize it with, like, one of these LLMs. So, that, you know, is like a very small use case. But when we have time and there's a need, that's something that we can definitely step in to do.
And do you think in the future you'll get, I don't know, you know, you could imagine doing predictive modeling for all sorts of kind of business aspects of the company, right? And that you might be the person with the skills most closely aligned with those goals, even though that's not really what your primary job is. Is that something you kind of anticipate being pulled into or maybe not? This hasn't come up yet, I guess. It hasn't come up yet.
I know that, you know, there are, so, for example, a lot of analyses go into trying to find what is the next right indication to go into among, like, a choice list, right? So, let's say you have these 10 different cancers and you think that your drug could work in all of them. But what is the right market to go into? And so, you know, I've seen our BD folks work on those kinds of analyses.
But that seems to me like it's very manual at the moment still, right? So, you have to do the analysis in terms of, you know, what is the patient population to begin with? And then what are the standard of care lines of therapies? And then after you whittle all of that down, you say, well, this is the patient population that we would be able to target after you account for all of these things. And then trying to figure out, you know, what is the right market in terms of the size of the patient, like the treatable patient population that you go into. But I've only seen the output of those kinds of analyses. I haven't seen what goes into it and sort of where they pull those data from. But it sounds to me like that's probably very manual, like a manual search kind of thing. You mean like it has to be manual or it just is currently manual? It has been. Yeah, it has been.
R and Python in the workflow
I see Bill asked, it was a follow-up question when we were talking about algorithms. And Bill had asked, what role does R and Python play in your workflow? We're all Python here. I was all R for many, many years. I think it really depends on the kind of analysis that you're doing.
So my personal preference is, you know, when we're writing pipelines and doing data processing, that's all done in Python. I'm really familiar with R in terms of the data analysis side and making figures and reports and things. So I really love Posit for that. You've been using it for a very long time. But it's our day-to-day, right? Every day we open up notebooks and do these kinds of analyses, whether in Jupyter or R.
Transitioning into bioinformatics
Jazz, I want to make sure your question, it may have been answered in another one, but if not, I want to make sure you have an opportunity to jump in. I think it was about roles to suggest for someone with, let me, and you can jump in if you want to add more context, but it was what roles can someone suggest for a life science business analyst with a math and biomedical science degree to make a transition into bioinformatics?
Jazz, I see you've unmuted. Do you want to jump in? Yeah, so I have 10 years of experience in healthcare analytics. So it has been somewhat business analytics, but it's also been healthcare information. But I see bioinformatics more towards the science side of things. So, you know, making use of, say, biomedical science knowledge, but I haven't really been doing that. So I just wondered how easy would it be for someone to jump away from business analytics into more bioinformatics and what kind of roles, because I think, I mean, it's easy to sort of just put in bioinformatics analyst, but I find that they require MSc. So are there any other roles that I could make that transition?
Yeah, I mean, I think it really depends on your current skill set and where you could jump in and sort of start working, essentially, from day one. So, you know, if you are already familiar with, you know, R or Python to run some analyses, you know, it doesn't really matter what kind of data are coming in, right? As long as you understand sort of what the distribution looks like, you know, what does the data represent? Those are all things that you can learn.
I think it's really making the case for whatever role you're looking into. I would assume if you don't have like a master's, you're probably looking at starting as a research associate level to make the case that you would be able to pick up these skills, given what you're already, like what you have already been able to do.
You know, I think that you probably have a better chance. Well, it's actually hard to say. You know, I want to say that you probably have a better chance at a smaller company, but it really depends on the company and what they're looking for in a role. You know, I think in a larger company, maybe you have a little bit more leeway in terms
of, you know, having like having the opportunity to learn skills, like a little bit more time to learn skills on the job, whereas smaller companies want you to really be able to hit the ground running on day one. But I think that probably also varies by company and role.
So it's really like trying to find the right match, like the right role for you, given your background, what you want and what the company needs.
Yeah, thank you. That answers the question. Thank you.
Career advice
I know we've had a few different career type questions here, so I want to make sure I ask this as well. Is there a specific piece of career advice that you've either given to other people or that you've received that stands out in your mind?
Yeah, I mean, so I think personally speaking, the most important thing to me is to not be afraid to make the leap into something new. Every time I started, I have started a new job, I have been terrified that it's not going to work out or I'm not going to have the skills. This happened when I was going to grad school. It happened when I moved my postdoc. It happened when I was, you know, looking for my new job and my second job and my third job every single time.
I think this is where it's really important to know for yourself internally what you think you're capable of doing. Right, so yes, you have to convince someone else to hire you that you're capable of doing what you say, but you also have to know that. And so I think once you are comfortable with that, it makes it, you know, then you don't fear as much the unknown of what the new job is going to look like. It's more that you know that it's going to be, you know, challenging for a couple of weeks or a couple of months as you start up, but that you're going to get there and it's going to be fine.
Yeah, I think fear is natural. Being nervous is natural, but I think that, you know, people are able to overcome that.
Yeah, I mean, so I think personally speaking, the most important thing to me is to not be afraid to make the leap into something new.
I love that. Thank you very much.
The five-year goal story
Something that you said to me also, which was like sticking with me is a boss said to you before, in an ideal world, what would you have accomplished five years from now in an interview? And I thought that was, it's just, I've just been thinking a lot about it too. So I was wondering if you could share that story with everybody.
Yeah, so I was, so this was for my very first job. I'd never had a, like a job job before. So I was a postdoc. I was interviewing. I was really excited for this opportunity because in my postdoc, I had been studying the microbiome and then I had this opportunity through a recruiter to work at a microbiome therapeutics company. I knew zero about drug discovery. I knew nothing.
So, you know, the, the, my hiring manager who later became my boss, she asked me like the last question before the end of the interviews, she was like, well, in an ideal world, you know, what would you like to have accomplished five years from now? And I like, I really hate that question because like, you know, what do you, what do you know? What, what's life going to look like in five years? Nobody knows.
So I was thinking to myself, I was like, well, you know, what, what, what would be really cool is if I could be part of the team that brought the first like microbiome therapeutic to the market. I had no idea how unlikely of a, of a, of a goal that could be. I didn't know what timelines look like in drug discovery. And, you know, five years is not a, not a very long time. And we were working on a novel therapeutic and a novel space, nothing like no microbiome therapeutic had been approved before.
But I, you know, what I like about this story is it kind of showcases, you know, when you are new to an industry and you don't necessarily know what to expect, that you can kind of aim for anything and, you know, not be held back by your preconceived notions of what's possible. And it, as it turned out, like five years later, after being at that company, five and a half years later, we were actually on track to, to FDA approval for that therapeutic. It's really amazing.
It's like kind of an incredible Cinderella story, which I love to tell because actually I'm going to add a little bit to that, to that story, which is when I started that job, I started on Monday and on the Friday of my first week and my first job ever, we got this really, really terrible news that our phase two trial had completely flopped. It was for the same program.
And you can feel like the, the disappointment and the heartbreak was really palpable in the room, right? Everybody had worked so hard on this program and the results were so bad. There was like nothing. There's no signal at all. And I thought I was like going to not have a job on Monday, basically.
But somehow really, I think, attesting to the scientific leadership, you know, everyone really, really dug down deep and I was fortunate to be a part of this effort. We went back, looked at the data, reanalyzed the data, built new algorithms, you know, built new software, got new data to, to really figure out what happened to identify all of the issues that led to that failed phase two trial. And we didn't, when we went back to the FDA, we didn't have to rerun the phase two trial. We justified like what we, you know, what we had found and what, why we thought we saw what we did in the phase two and we were able to proceed to a phase three trial. It was really an amazing story.
But I, you know, what I like about this story is it kind of showcases, you know, when you are new to an industry and you don't necessarily know what to expect, that you can kind of aim for anything and, you know, not be held back by your preconceived notions of what's possible.
Microbiome therapy and the pill
I know we have a few minutes left here. I want to make sure I get to all the questions. Matthew, I see your hands raised. Want to jump in? You might be muted still. Okay, I'll jump over to another one, but I see your hand still up. If you unmute, feel free to go next.
Let's see. Oh, Nathan had a question just about the microbiome therapy and just more details on that, if that was the same as fecal transplant. Different. It's actually an oral pill.
So FMT, this fecal microbiota transplantation, this is something that has existed for, for several years, for a long time for the treatment of recurrent C. diff. This was a procedure that was not regulated by the FDA, but it is like very efficacious for the treatment of C. diff infections. So what Ceres was trying to do was to make this kind of a regulated, a clean, clean process, because an FMT is really just like a fecal enema, which nobody really wants to have done.
And so what Ceres was developing was this pill, which was derived, which is derived from healthy patient stool samples that has been heavily, heavily processed, so that all that remains are these bacterial spores from commensal bacteria. And so I've actually seen those pills that are like, you know, just like a normal pill, little blue pill. And it's just packed full of healthy bacterial spores.
And as, as, you know, one of the scientists who got to analyze the clinical data, what you see is when patients take this pill, they start out with a very dysbiotic microbiome, because this is, these are patients who are really sick. They've taken lots of antibiotics. Those antibiotics have really destroyed their gut microbiome. There's nothing there. And within like two to three days of taking this pill, the microbiome is just like, it's blossoming back. It's really, it's really incredible to see the kinetics.
AI and LLMs in drug development
Thank you. Sol, I see you had a question in the chat just a few minutes ago. Wanna jump in next? And it was on LLMs or generative. Yeah, can you guys, oh, sorry, Rachel. Oh, yeah, sorry about that. Yeah, I think, you know, it's interesting, Liang, that you mentioned like the transition of like bioinformatics to data science, but with like all the rise in AI and stuff like that, I was just interested, you know, has your team found any interesting applications for LLM or other generative AI products recently? We'd love to hear about that.
Not my team. So I've actually been to a couple of conferences slash workshops where people are, you know, throwing around this generative AI idea in the drug development space. And what I have found is that in general, there's a lot of hype and not a lot of results, specifically when you're talking about gen AI, right? And I also think that over time, what people call what I would call just statistics and analysis has changed. So, you know, seven years ago, it was, what are we doing with machine learning? This, you know, someone talking to me, right? And then, you know, five years ago, it's like, what are we doing with, you know, deep learning? And now it's like, what are we doing with AI?
And I think that for the kinds of analyses we do, the technology is not there yet for that. I think it's different in the chemistry space where you have a lot of data points going in and you have a lot of screens being run. So you have something that you can run on, that you can generate these models on. When you're talking about like microbiome patient data in like any kind of clinical trial, honestly, the sample sizes aren't large enough to build those kinds of models. And I don't think that people have a clear idea of really of even what are the right questions to ask with those kinds of models. That's my, sorry, it's kind of a downer, but that's my take on it right now.
You know, I was speaking to our co-founder a couple of weeks ago and he was laughing. He told me that, you know, he had spoken to this investor recently and she was like, yeah, you know, half of these companies that are touting their AI capabilities, like you can do what they do in Excel. It's not what you think it is. That's funny. I appreciate that perspective. It's actually incredibly useful to get like the, you know, the real kind of perspective. So thank you.
FDA submissions and reproducibility
Thank you. I think we have gotten to every question so far. Travis, you told me to put this one at the bottom of the priority list. So you get to go right now. Thanks. I was curious, you know, there's a lot of scuttle and industry around how do we do FDA submissions in R only and, you know, for standard clinical trials, there's a lot of movement there. I'm grateful that other companies are kind of taking lead there. But what you at Roam do is a step beyond really. And hearing that you're using Python maybe makes it an extra step of challenging, like validating Python environments comes to mind. I didn't know Kevin was here. He probably might also have commentary on this topic. But yeah, what does that look like for you? Have you had to manage that? Like, how do you validate and ensure that the FDA or other regulators believe in the novel bioinformatics pipeline that you're implementing?
It's a really good question. We actually struggled with that quite a lot when I was at Ceres and we were preparing our BLA filing. And so, I mean, I don't think I have a really good answer for that. We do the best we can to perform the kind of validation experiments that we think need to be done to validate those environments. I mean, Kevin may know better, you know, being from the FDA. But, you know, we captured all of our environment variables. We were using R at the time. So, you know, there's a RM package, right? So, we used that. We used all of the packages that we could to make sure that our environment was unpackable and reproducible by anyone.
And that was really, I think at the time, we were also considering like Docker images. We were trying to, you know, we were wondering like what kind of data do we need to submit to the FDA? You probably don't want all of our FASTQ files, like our raw sequencing data. You probably don't care about that. But like at what level, what is the right level of data to submit? And we had a lot of internal struggles trying to define that as well.
You're working with CBER? Sorry? You're working with CBER, the Center for Biologics, I'm guessing? I don't know. Okay. Yeah, I don't know. Yeah.
Did you use biocompute objects? Are you familiar? Okay. Are you familiar with what they are? Okay. It's a, it's something that the FDA developed as a, like a format for storing all of your parameters in your pipeline, you know, in a file. So that's, it's a, you can kind of send this file and it makes your code reusable, you know, or. Okay. Yeah, it makes things reproducible.
I'm not, I wasn't personally involved very much in it. It's been a CBER thing, you know, because they dealt more with this. Yeah. But anyway, that's interesting. I would like to connect with you, actually, after this to talk about this more, because we will have more of these kinds of issues coming up as we're also moving into the clinic here at Rome. Okay. Well, yeah, and I'm not in. Could I be able to fly in the wall too? Yeah. Right? Let's spin off the.
One nerdy question is this, how did you deal with human contamination in the reference genomes? Because I did, yeah, I've run into that problem.
I mean, we do, we do the double mapping. So first for the microbiome stuff, we map to the human reference genome first and remove all of that. In some indications, it's like much more critical than others. So we also had studies in ulcerative colitis and in the patient's stool samples that were, you know, that had a lot of blood contamination, for example, you can see that 95% of your sequence is coming from human data.
And well, and the, did you ever stumble across the FDA Argus database? That was a project from CDRH where they're trying to make reference grade genomes for basically clean, clean microbial genomes as references available. But it sounds like maybe, maybe not. Okay.
All right. I think we're out of time. I know we're a little bit over. It sounds like this is a really fun conversation for everybody to have. And I would love to listen to that too. Kevin sounds like we're meeting a FDA submissions chat as well.
Yeah, actually. Thank you so much, Leanne. I really appreciate you taking the time to join us today and sharing your experience. This has been great. Yeah. Thank you, everyone.