Latest / The BioHub - by Avetix / Episode 247 - Chris Moxham Founder & CEO and Clayton Mellina CTO of Transcripta Bio

Transcript

Episode 247 - Chris Moxham Founder & CEO and Clayton Mellina CTO of Transcripta Bio · The BioHub - by Avetix

• 0:06
  Today I'm joined by Doctor Chris Moxon, Co founder and CEO and
• 0:10
  chief scientific officer of Transcriptor Bio, alongside the
• 0:14
  company's chief technology officer, Clayton Molina.
• 0:17
  Together, they are combining transcriptomics, high throughput
• 0:21
  biology and AI to transform the way new medicines are
• 0:24
  discovered. Chris Clayton, welcome to the
• 0:26
  Bio podcast. Thank you for having us
• 0:29
  Likewise. And to start, could you both
• 0:31
  tell us a little bit about your background and what led you to
• 0:34
  founding Transcripter and maybe Chris, you could start us off?
• 0:38
  Sure. I've been in drug discovery
• 0:41
  pharma for 30 years now. I'm a pharmacologist by
• 0:44
  training, spent 20 years at Eli Lilly and Company, as I like to
• 0:48
  say, getting a really good education in drug hunting, have
• 0:52
  a particular knack for building data-driven platforms that you
• 0:56
  know, has led to over 15 molecules going into the clinic
• 1:00
  across multiple different therapeutic areas.
• 1:02
  And after my time at Lilly, I was the CSO at Fulcrum
• 1:05
  Therapeutics and helped take a couple programs into the clinic
• 1:08
  there. And the reason to and the
• 1:11
  impetus to found transcriptive bio really was a confluence of
• 1:14
  technology and the ability to interrogate the transcriptome at
• 1:18
  scale as well as the ability to to combine phenotypic drug
• 1:22
  discovery with target based drug discovery to to find innovative
• 1:25
  answers for patients a lot faster and a lot cheaper.
• 1:28
  And Clayton, can you tell me a little bit more about your
• 1:30
  background Prior to joining transcriptor?
• 1:33
  Certainly. So my background is in software
• 1:36
  machine, software engineering and machine learning.
• 1:41
  That's sort of my formal training and also my career
• 1:43
  experience up to joining Transcriptor.
• 1:46
  Previously I was working not in life sciences on on.
• 1:52
  I would characterize them as sort of generic computer vision
• 1:56
  applications in industry, things like virtual and visual
• 2:01
  inspection in manufacturing, assembly lines, product search,
• 2:06
  understanding logistical computer vision, these sort of
• 2:10
  applications. I came to the life sciences
• 2:14
  because I was diagnosed with a rare disease myself.
• 2:17
  And so in the process of learning about it, educating
• 2:20
  myself as a patient, I sort of realized I sort of slept on
• 2:24
  biology, so to speak, in in my educational history.
• 2:28
  I found it extremely fascinating to learn about and it became
• 2:31
  clear to me that it was kind of a frontier of applied
• 2:37
  computation that was really exciting just getting started
• 2:41
  and which had a lot of direct positive impact.
• 2:46
  So I got interested. That helped organize some
• 2:48
  hackathons that brought together clinicians and data scientists
• 2:51
  friendship for instances like 8 years ago or so, and kind of the
• 2:55
  early days when that wasn't so common.
• 2:57
  A combination of of talents together and sort of it was a
• 3:02
  slow transformation into my current role, basically
• 3:07
  bringing, you know, computational technology,
• 3:09
  machine learning, AI to the drug discovery process at Transcripta
• 3:14
  well. Thank you for sharing your
• 3:16
  backgrounds and Chris, for the listeners who are hearing about
• 3:20
  Transcripta for the first time, what is the company building and
• 3:23
  what problem are you setting out to solve?
• 3:26
  Yeah, I'll start with the problem that we're trying to
• 3:27
  solve first. We all understand that the
• 3:32
  dilemma in in pharmaceutical R&D is that many times you fail in a
• 3:36
  phase two clinical trial when you're trying to demonstrate the
• 3:38
  proof of concept. And then of course, that failure
• 3:42
  comes after years of investment of time and 10s of millions of
• 3:46
  dollars in capital. And of course, that failure
• 3:50
  means that patients don't get an effective therapy.
• 3:53
  The company fails because they've made a choice to go down
• 3:56
  the road and spend all that time and money, but also the
• 3:58
  opportunity cost of other things they could have done.
• 4:01
  And that's why the cost of getting a drug approved today
• 4:04
  is, you know, close to $2 billion.
• 4:06
  So we're trying to solve that problem, that is to increase the
• 4:09
  success rate in phase two clinical trials.
• 4:13
  And we go about that by pulling forward all of the de risking
• 4:16
  that we can. So if you think about different
• 4:19
  failure modes, well, one primary failure mode is that you pick
• 4:22
  the wrong target to start. And so our solution to that is,
• 4:26
  well, let's go to the places where Human Genetics tell us
• 4:28
  unequivocally this gene or this process causes that disease, the
• 4:33
  root cause. Let's focus on the root cause of
• 4:35
  disease. The second perhaps is that, you
• 4:38
  know, scientists have spent a lot of time trying to build
• 4:40
  models in the lab thinking that that can recapitulate disease
• 4:45
  biology in a human. Our approach is to actually
• 4:48
  start with data directly from patients.
• 4:50
  And so we ingest single cell RNA, seek data directly from
• 4:52
  patients to build these disease signatures that I'll let Clayton
• 4:56
  talk about. Thirdly, we do all of our
• 5:00
  screening of of small molecules in cell based assays.
• 5:05
  And so rather than you know the traditional approach where you
• 5:07
  start with a binding of a molecule to a target maybe
• 5:12
  weekly, you don't understand a lot of other things.
• 5:14
  We're actually starting from much more mature position one,
• 5:18
  because we have cell based activity, we understand the
• 5:20
  selectivity but also the types of molecules that we screen.
• 5:23
  We know that these are going binding to druggable targets and
• 5:27
  so we're essentially starting at lead optimization after we
• 5:29
  complete a screen. And then lastly, you know the
• 5:32
  traditional approaches, one always starts from a de Novo
• 5:35
  position and that takes you know a fair amount of capital, $25
• 5:40
  million typically to go from a screen to an Ind and then you're
• 5:44
  just getting started. Our approach in terms of the
• 5:46
  pipeline that we're building is a tiered approach.
• 5:49
  So again the big dilemma is capital and efficiency.
• 5:52
  Our approach is to make a much more capital efficient approach
• 5:54
  by having a tiered pipeline of assets that yes, there's a novel
• 5:58
  molecules in there. There are clinical stage assets
• 6:01
  that have already been de risked in a significant way.
• 6:04
  And there are even approaches where we're we've identified FDA
• 6:07
  approved drugs that can be repurposed into a new indication
• 6:11
  and all of that is meant to create a much more capital
• 6:13
  efficient company versus the traditional approach.
• 6:17
  That's a really interesting approach there as well.
• 6:20
  And Clayton, then drug discovery has traditionally been centred
• 6:24
  around identifying and validating targets, as Chris
• 6:27
  just mentioned. How does your approach differ
• 6:30
  with that? Yeah.
• 6:31
  So we take what you would characterise as a more
• 6:34
  phenotypic approach, which is, you know, we use omics data to
• 6:43
  understand what cellular functions are dysregulated in
• 6:47
  disease states. Our computational platform does
• 6:51
  this across hundreds of diseases, hundreds of cell
• 6:55
  types. And so we get a rich
• 6:57
  characterization of, you know, what are the effects of diseases
• 7:00
  or even disease subtypes on particular cells and their and
• 7:03
  their processes. This is important for us because
• 7:09
  those cellular processes are really what cascade downstream
• 7:13
  into the disease symptomatology. And so they represent sort of
• 7:16
  the cellular dysfunction that's at the root cause of diseases,
• 7:20
  but they also represent coordinated changes in the in
• 7:24
  the cellular activity that could involve multiple genes or
• 7:29
  pathways that are being dysregulated.
• 7:32
  And so that allows us to sort of cast a wide net on potential
• 7:36
  therapeutics, right? It allows us to identify drugs
• 7:41
  that modulate those root cause processes in beneficial ways for
• 7:46
  the disease before actually committing for instance, to a
• 7:51
  single protein target, say that we're trying to design A
• 7:55
  molecule to bind to. It's not to say that those
• 7:58
  target driven approaches aren't useful, but we think that a
• 8:02
  large kind of mistake of the field has been in rushing, if
• 8:06
  you will, to reduce the drug discovery problem to a
• 8:11
  particular target binding problem too early.
• 8:14
  And so in some sense, we leave open the potential of target
• 8:20
  identification much later in the process and also thereby kind of
• 8:25
  widen the net for potential therapeutic strategies in a
• 8:29
  particular disease. Yeah.
• 8:31
  If I could just add real quickly, I think what Clayton
• 8:33
  touched upon is that in the very beginning of the process, right,
• 8:37
  we capture the benefits of phenotypic screening, casting a
• 8:40
  wide net in an unbiased way to identify the best opportunities
• 8:44
  to modulate this disease signature that comes from
• 8:46
  patients. But then we can transition
• 8:48
  because of the types of molecules we screen to a very
• 8:50
  efficient target based drug discovery approach.
• 8:52
  And so it's the synergy of those two things that come together
• 8:55
  that really give us an unfair competitive advantage.
• 8:58
  Yeah, that's a really fascinating approach.
• 9:01
  And one of the concepts behind your platform is matching drug
• 9:05
  signatures with disease signatures.
• 9:08
  Clayton, can you explain what that means in practice and why
• 9:12
  it's such a powerful way to identify these therapies?
• 9:15
  Sure. I just described a a bit of that
• 9:18
  pillar, that disease Atlas as we call it, which is so this
• 9:22
  repository that's the output of our processing pipelines of
• 9:26
  disease signatures that characterize cellular
• 9:30
  dysfunctions. You know, the, the, the
• 9:33
  complementary pillar of the company is what we call the drug
• 9:37
  gene Atlas, which is our repository of proprietary
• 9:41
  transcriptomic drug screen data. This is a, a data set we, you
• 9:46
  know, we, we run these large scale screens here in our lab in
• 9:49
  Palo Alto and they deeply characterize the effect drugs
• 9:54
  have on particular cellular systems, often at multiple doses
• 10:00
  of the drugs. We understand causally how
• 10:04
  changes in that drug dose Dr. downstream cellular function.
• 10:09
  And so in with these two sort of pillars, we we basically find
• 10:14
  commonalities or complementarities between these
• 10:18
  pillars, right. So you know, we may observe an
• 10:22
  interesting cellular function or maybe pathway disruption in a
• 10:26
  particular cell type in a particular disease, right.
• 10:29
  We can ask, OK, is that a pathway compensated by a drug
• 10:34
  perturbation? That's that's apparent in our in
• 10:37
  our drug screen data. That's a kind of a simple
• 10:40
  example of how we do these, these matches, You know, I
• 10:44
  would, I characterize, you know, perturbations drugs and diseases
• 10:47
  as biologically dense. So typically a drug is doing
• 10:50
  multiple things, maybe some things that aren't necessarily
• 10:57
  the core reason why that drug is indicated in or being studied in
• 11:02
  a disease that it is, but is an auxiliary effect that actually
• 11:07
  is useful in another disease area.
• 11:10
  And so that matching process is exactly that, understanding the
• 11:13
  common biology of drugs and diseases at scale to find
• 11:18
  opportunities to start new programs with those underlying
• 11:21
  biological insights. And with that, then what type of
• 11:26
  insights has that approach enabled?
• 11:28
  That would have been difficult using more conventional methods.
• 11:32
  I think a good example is our autism project.
• 11:37
  We so, you know, to just relate the story briefly, we several
• 11:43
  years ago began began studying a rare autism spectrum disorder
• 11:50
  that was genetically caused. But we sort of characterized the
• 11:56
  phenotypic state that was downstream that genetic insult
• 12:00
  in neurons. And through this process that we
• 12:05
  ran here in the lab, you know, discovered a class of drugs that
• 12:10
  regulates hyper excitability phenotype in that that patient's
• 12:15
  neurons. This class of drugs have been
• 12:18
  developed for cancer in fact, but we discovered that a much
• 12:23
  lower dose of the drug than you would use for its cancer
• 12:27
  indication was effective at regulating sort of neural
• 12:32
  processes that we believed lead downstream to neurodevelopmental
• 12:38
  challenges for this patient. That was a a single patient
• 12:42
  cohort at the time. We took that drug through
• 12:47
  clinical trials with that patient saw behavioural.
• 12:51
  It's improved and expanded the scope of the program recently by
• 13:00
  identifying a similar transcriptomic signature in a
• 13:05
  broader class of genetically defined autism spectrum
• 13:10
  disorders. And so this was a really
• 13:12
  interesting example where we have a signature that
• 13:16
  characterizes the disease state at the transcriptomic level.
• 13:21
  And the the ability to recognise that that disease state is
• 13:25
  actually converges in a large number of different autism
• 13:30
  spectrum diseases, which it currently, you know, is vastly
• 13:34
  expanding our program to, to sort of apply the same drug
• 13:39
  class potentially to treat a wider range of, of diseases.
• 13:45
  So that's I think a very powerful example, which is very
• 13:48
  unique to our our platform. Absolutely.
• 13:51
  I think another one that really captures the speed at which we
• 13:55
  can move is our focus in Huntington's disease.
• 13:58
  And this is where we're targeting the underlying
• 14:01
  mechanism that is known to 'cause these repeat expansions
• 14:04
  that ultimately lead to the onset of Huntington's disease.
• 14:08
  And just last year a paper came out in Science in February
• 14:12
  demonstrating that if you knock down MSH 3, which is a critical
• 14:15
  component component of this Mute S beta complex, you can delay
• 14:19
  the repeat expansion in in patient cell lines.
• 14:22
  Queried our data set and within minutes we had identified novel
• 14:25
  structures that we were in fact working on that could down
• 14:28
  regulate MSH 3 mRNA and protein. And today is now a program that
• 14:33
  is rapidly moving for towards a development candidate and that
• 14:36
  we hope to take into the clinic next year.
• 14:38
  So it's just another feature of the data that we've amassed as
• 14:42
  proprietary data set that I don't think anybody else has
• 14:45
  that allows us to act with either such innovation and
• 14:49
  insight to try to sort of kick down the door in areas that
• 14:52
  really require it as well as to move with unprecedented speed.
• 14:57
  Yeah, that's really fascinating to see the progress that you're
• 15:00
  having there with with that platform.
• 15:02
  And Chris, back to you and you are announcing A partnership.
• 15:06
  Can you tell the listeners a little bit more about that and
• 15:09
  what it means for transcriptor? Yeah.
• 15:12
  So this is a partnership with the Mayo Clinic and we'll also
• 15:17
  be announcing that they've invested in the company as well
• 15:20
  in transcript a bio, which is great.
• 15:22
  I think as everybody understands, the Mayo Clinic is
• 15:25
  a world renowned medical institution and plays such a
• 15:29
  critical role in patient care for the last 150 years.
• 15:33
  And we are truly privileged to be entering into research
• 15:37
  collaborations and partnerships with them in two distinct areas
• 15:41
  so far. 1 is in autism spectrum disorder, where we will partner
• 15:46
  with clinicians at the Mayo Clinic to help us identify
• 15:49
  additional patients who have the signature that Clayton was
• 15:52
  referring to and also with the intent to have the Mayo Clinic
• 15:57
  be a a clinic, a site for our clinical trial that we intend to
• 16:00
  start next year. The other is in the area of ALS
• 16:05
  and so here the Mayo Clinic has amassed again tremendous
• 16:10
  biobanks of patient samples. And so we are going to gain
• 16:14
  access to those samples to essentially do what Clayton just
• 16:16
  described. Can we create single cell RNEC
• 16:21
  data sets from tissues from the brain and spinal cord from ALS
• 16:24
  patients? Can we elaborate a disease
• 16:26
  signature and then run through that process of our platform
• 16:29
  that is can is there something in our drug gene Atlas that
• 16:33
  perhaps reverses that disease signature that then becomes the
• 16:36
  jumping off point for a drug discovery effort?
• 16:39
  Of course, the other piece that the the Mayo Clinic brings to
• 16:42
  the table is that they've done a tremendous job of prospectively
• 16:45
  capturing data from patients to understand disease progression
• 16:48
  and how to differentiate. And so again, we are beyond
• 16:51
  thrilled and privileged to be entering into this partnership
• 16:54
  with the Mayo Clinic, and I expect that there'll be many
• 16:56
  more to come. Well, a big Congrats on that and
• 16:59
  glad that you're discussing that on the podcast today as well.
• 17:04
  And then maybe back to the AI part of of your business, it's
• 17:08
  evolving incredibly quickly. How are you applying AI within
• 17:12
  Transcript, the Transcript platform and where do you see
• 17:16
  the biggest opportunities for it in the future, Clayton?
• 17:20
  Yes, it's certainly moving quickly and it's been
• 17:24
  remarkably, you know, such a remarkable resource at
• 17:28
  Transcripta. You know, we started with our
• 17:32
  mainly focusing on perturbation prediction type problems.
• 17:36
  So in the, in the field, this is sort of the term of art for, you
• 17:40
  know, asking, can you train a machine learning model to
• 17:43
  predict what will a particular perturbation like a candidate
• 17:47
  drug structure do if it were introduced to a cellular system?
• 17:53
  And for us, that's predicting, you know, changes in gene
• 17:57
  regulation events and, and this has enabled us to do several
• 18:02
  successful virtual screens. So I think recently with the
• 18:07
  advent of AI agents, the scope of, of utility for AI is just
• 18:12
  massively increased in, in science in general, right?
• 18:16
  For us, you know, it's, it's extremely useful at scaled
• 18:22
  interpretation of transcriptomic data.
• 18:27
  Transcriptomic data is complex. You're measuring 20,000 gene
• 18:33
  regulation events, you know, often in dose response format in
• 18:37
  multiple cell types. It's very challenging for a
• 18:41
  human to synthesize and understand what's happening in
• 18:44
  terms of cellular function from that data.
• 18:47
  There's tons of traditional tools to do that.
• 18:50
  Certainly things like pathway enrichment analysis is
• 18:53
  important, you know, methods that allow you to ask what sets
• 18:56
  of genes are coordinating in a in a cell to achieve some
• 19:00
  function and how are those functions changing.
• 19:03
  But AI, especially AI agents really enables a new sort of
• 19:06
  type of interpretation of this high content omics data.
• 19:12
  You know, First off, it allows the AI agents are able to dive
• 19:16
  deeply into particular transcriptomic changes and
• 19:20
  understand, relate things what's changing to, to, to one another
• 19:24
  and, and referring to prior literature.
• 19:27
  And so giving, you know, gives a new interpretive lens on that.
• 19:32
  And I think it also enables a company like us to scale that
• 19:36
  activity really broadly, right? Having agent swarms interrogate
• 19:41
  different hypothesis starting points for, you know, disease
• 19:46
  program initiation and building evidence incrementally over time
• 19:53
  in in a way that a small human team wouldn't be able to for
• 19:56
  this type of data. So I think, you know, these
• 19:59
  things are really just getting started and we're experimentally
• 20:03
  experimenting massively with them, really happy with how
• 20:06
  they've changed our process recently and certainly expect a
• 20:10
  lot more changes in in the future.
• 20:13
  Yeah. I think this comes on well to
• 20:16
  our next question. I think looking ahead, what do
• 20:19
  you think drug discovery will look like 5 years from now?
• 20:22
  And maybe what role will AI play in getting us there?
• 20:26
  Yeah, I mean, I think one way or another, AI in general will
• 20:32
  radically transform how we do science and drug discovery is is
• 20:36
  certainly no exception. I think, you know, the ability
• 20:42
  to maintain as open possibilities thousands of
• 20:48
  hypotheses for how disease works, diseases work, how drugs
• 20:52
  work and potentials potential for new biology that can
• 20:58
  underlie drug programs. That's something that it's hard
• 21:01
  for even a, a large pharma to do, right, to have actually
• 21:06
  thousands of, of promising hypotheses sort of currently
• 21:10
  open and under investigation. But really I think that's what
• 21:15
  AI is able to do. And also, you know, allow small
• 21:20
  groups of human scientists to sort of orchestrate that type of
• 21:24
  activity. So it changes what science is
• 21:27
  day-to-day, but it also will enable a lot better ideas to
• 21:33
  come to the surface and drive the experiments that we do,
• 21:37
  accrue evidence and ultimately lead to much better drugs
• 21:41
  reaching clinic. Yeah.
• 21:43
  I mean, I think just building on that, I think the scale at which
• 21:48
  even small companies like ours will be able to operate, the
• 21:52
  ability to compute and then interpret from that, but then
• 21:55
  also the intelligence to execute, right?
• 22:00
  And big pharma companies, many times that institutional
• 22:03
  knowledge is captured in a, a smaller subset, right, of the
• 22:06
  overall company and that institutional learning is, is
• 22:10
  carried on. But but the, with the
• 22:12
  introduction of AI and these agents, that's no longer the
• 22:15
  case, right? I think the, the expert, the
• 22:17
  true expertise, right? And the ability to make
• 22:19
  judgement calls is going to be scaled and distributed.
• 22:23
  And so ultimately for patients, I think it's going to be more
• 22:26
  effective drugs more often and I think it's going to be cheaper
• 22:30
  to conduct drug discovery as well.
• 22:32
  Yeah. And hopefully that is the case
• 22:34
  with the market, right? Well, thanks so much, guys.
• 22:38
  Just to follow off, finish off with the last question.
• 22:42
  We asked all of our guests on the Biohub podcast.
• 22:44
  If you could choose any scientist or innovative past or
• 22:48
  present to join you for dinner, who would it be and why?
• 22:51
  And Chris, maybe you could start it off.
• 22:54
  Guess mine would be Thomas Edison.
• 22:58
  It's not just enough to have an invention, right?
• 23:02
  Innovation really means that it has to lead to an impact.
• 23:05
  And that's really been the fundamental basis of my sort of
• 23:08
  ethos, if you will, as a scientist and a drug hunter in
• 23:11
  my career. I'm not just interested in
• 23:13
  working on some interesting invention, but rather something
• 23:17
  that will lead to an impact in patients.
• 23:20
  So much more combination of innovation and and pragmatism to
• 23:24
  what really matters and what to what's going to really move the
• 23:27
  needle. And so I think picking Thomas
• 23:30
  Edison's brain on how to best do that would be pretty
• 23:32
  interesting. A.
• 23:33
  Great choice. And what about you, Clayton?
• 23:36
  Yeah, I think mine would be John von Neumann.
• 23:41
  You know, as a his his contributions have been are so
• 23:46
  wide, but all anchor in, you know, basic, a basic picture of
• 23:54
  computation and how that plays a role in a variety of natural
• 23:58
  processes. I think his his work on self
• 24:02
  replication, for instance, was, you know, it's foundational in
• 24:07
  connecting ideas of computation to biology.
• 24:11
  And I think, you know, as a theoretical program, it's still
• 24:15
  sort of being played out in a lot of ways.
• 24:18
  So I think his influence in scope and that sort of
• 24:22
  recognition of of the overlap of biology and computation to me is
• 24:27
  so fascinating and and so impactful.
• 24:30
  That's an interesting one as well.
• 24:32
  Well, thank you both for joining us today.
• 24:34
  It was great to learn a little bit more about transcriptor, and
• 24:37
  good luck with what you're building as well.
• 24:40
  Thank you so much for having us. Thank you, it was a pleasure.