Scroll Top

Raphael Townshend on The Power of Small Molecule Drugs

Episode Highlights

  • A small molecule drug is a chemically synthesized molecule with a low molecular weight, typically less than 900 Da, that interacts with specific molecular targets in the body to achieve a therapeutic effect..
  • ARES (Algorithm for RNA Structural Exploration) is a machine learning algorithm developed by Atomic that can predict RNA structure accuracy based on a few examples of known structures.
  • The algorithm focuses on the 3D space and the elements (atoms and their positions) within the RNA molecule to derive other information.
  • The relatively neglected nature of RNA drives atomic’s focus on RNA structures compared to proteins in the field of AI drug discovery.
  • Non-coding RNA plays various roles in cellular functioning and disease, and understanding its structure and function can lead to new insights and therapeutic opportunities.
  • Historically, predicting RNA structures and finding molecules that bind to them has been challenging. Still, Atomic’s approach aims to overcome these difficulties by leveraging the unique 3D structure of RNA and its functional implications.
  • PARSE (Platform for AI-driven RNA Structural Exploration) is an integrated wet lab and computational platform developed by Atomic, which includes ARES as its core structure engine.
  • PARSE combines AI algorithms with genomics-based assays to train and improve the prediction algorithm and guide further experimentation.
  • The goal of PARSE is not only to predict RNA structures accurately but also to identify structured and ligandable regions within the transcriptome, facilitating drug discovery targeting specific RNA molecules.
  • Atomic aims to expand its capabilities in machine learning on atoms and RNA, tackling problems related to predicting RNA structures, modeling flexible RNA molecules, and utilizing generative modeling technologies.

Full Transcript

Harry Glorikian: Raphael, welcome to the show. 

Raphael Townshend: Pleasure to be here. Thanks for having me, Harry. 

Harry Glorikian: Yeah, So, well, first of all, like, super congratulations on your funding news. I mean, I know you guys just announced on January 25th that you guys raised $35 million in a series A financing on top of the previous $7 Million in seed funding. Right. That is no small feat, for those people that are listening. Can you explain in broad terms. What the company’s mission is and what’s so special about your technology approach? 

Raphael Townshend: And at a high level we really see Atomic AI as sort of at the intersection of two of the largest breakthroughs of the last decade. I think everyone’s heard of the RNA revolution with the mRNA vaccines, things like that. But also the AI sort of revolution with all these new sort of cutting edge AI technologies. And really we see atomic as operating at the intersection of these two, starting with my own work, which for my PhD, essentially which ended up actually featured on the cover of Science, to really create this new field of RNA drug discovery with this cutting edge combination of RNA biology and AI technologies. 

Harry Glorikian: We’re going to come back to this a lot, in sort of my the way that I’m thinking about approaching this. But how does Atomic deliver on the vision you spelled out in the 2021 Science paper? 

Raphael Townshend: Right. So there’s been this sort of large amount of progress in AI, specifically for structural biology, which is sort of the area of biology which has to do with the shape of molecules. And really one of the central mantras of structural biology is that structure determines function, right? The shapes of things determine what they do, which is really almost too obvious in some ways, like the shape of a bike is important to the fact that it can transport you around. Right? If the wheels were in the wrong place, right, it wouldn’t do a very good job of that. And so very similarly, by understanding the shapes of the molecules, you can understand what they do and then also understand when things go wrong and then potentially intelligently design medicines to fix those things. And the progress on the AI side of things is that you can take any sort of molecule and really predict their shape overall, which is a process that typically can take months or years, if doable at all, and you’re bringing that down to seconds. 

Harry Glorikian: So, you know, I, I think the story of Atomic would make a lot of sense to listeners if we can talk a little bit about your personal journey, how the Science paper got published and how the company came together. I mean, along the way, we should talk about obviously the science of RNA and all of that will help us set up like how you are, you know, developing a drug discovery business. So first off, can you tell me the story of what you studied in grad school and how you got interested in machine learning to predict molecular structure? Because if I understand or have done my homework correctly, right, you were a PhD student in the laboratory of Stanford Professor Ron Dryer, who is a computer scientist and not a biologist, if that’s right. 

Raphael Townshend and Harry Glorikian on Small Molecule Drugs

Raphael Townshend: Yeah, I would describe him sort of as a computational structural biologist, at least somewhat. I would say he has, he definitely has very strong sort of biology knowledge, but from a very computational standpoint. Okay. To give you a little bit of background as to how we ended up here. There’s a long version and a short version, so I’ll try and stick with the short version. Okay. I actually started my PhD in the Artificial Intelligence Laboratory at Stanford, working on computer vision, actually working on 3D vision, self-driving cars, things like that. And really, there was sort of this realization that I came to that there was a lot of people that were trained very similarly to myself working on the same problems. And I realized that this tool that is AI could potentially have impact in much broader, in a much broader sense. And so maybe a simple way of describing is I kind of had this hammer, AI, right, in some ways that I’m going around looking for an area that is relatively neglected, right, in some sense in terms of not many people working in that way, but that was very high impact. And so there’s this rotation program at Stanford where you can rotate through different labs. And I essentially rotated through Ron’s lab. Did no AI, it was just trying to understand the space as best as possible. And after that rotation, I realized, hey, this is actually seems like a very good fit. There’s not many people working in structural biology. I feel like there’s a few in the world at the time, but if you can realize that you can really unlock a lot of new potential in drug discovery. And, you know, to make a long story short, then I had a seven year PhD where I had to learn a lot of structural biology, essentially, but that actually ended up being quite successful at the end of the day, leading to this work predicting the structure of RNA, using these cutting edge AI methods. And overall, in computer science it’s actually fairly common to go straight from a PhD to a faculty position from there. And so based on the success that’s where I thought I was going to go. But I started writing these grants around using this technology, and two things became very clear. The first one was that there was a large commercial opportunity around finding the shapes of these RNA molecules. And I realize  I’ve only described that at a high level so far. So definitely happy to dig into that specifically. And the second was really that you needed to scale the amount of data that you were collecting to train these AI models to a level that would be relatively inaccessible in an academic context. And so through those sort of two key realizations, essentially Atomic was born about two years ago, and maybe I’ll cut it off there for a second, but that’s a lot of the genesis of Atomic AI. 

Harry Glorikian: Well, but it’s interesting. I mean, you use the word rotation when you said it, and I was just thinking about, you know. ARES. which, you know, the acronym you give it right? But you develop this model called ARES, if I remember correctly, for Atomic Rotational Equivalent Scorer. We come up with so many good acronyms in our world. But and it turns out that it really performed incredibly well at predicting whether a predicted RNA structure was true to sort of real world data. In other words, much better than previous methodS. I mean, was that a surprise when you did this the first time? 

Raphael Townshend: Yes. The short answer is yes. You know, to dive a little bit more into that PhD, I actually started working on proteins and protein interactions, but the whole dream was really training machine learning methods on molecular structure systems of atoms or something like that, right. Ron, my advisor actually gave me a plaque at the end of the at the end of my PhD, that says atoms are everywhere. It’s actually it’s blurred out, but it’s actually sitting behind me right now. But the fundamental idea is proteins on a DNA, you know, carbohydrates, etc. They’re all made of atoms at the end of the day. And so if you design the right machine learning methods to work on systems of atoms, that should be fairly generalizable to new kinds of molecules. And so once we had our initial success on proteins, essentially, we then sort of approached this group at Stanford that were experts in RNA structure and said, Hey, we’ve got these methods that work well in proteins, let’s try them on RNA. And that worked amazingly well, almost immediately. And it was just very exciting at the time and really sort of highlighted to me how neglected in some ways this area was from a machine learning standpoint. If you could get that kind of amazing result so quickly, you know, who knows what else is out there essentially. 

What is a small molecule drug?

A small molecule drug is a type of drug that consists of small, chemically synthesized molecules with a low molecular weight (usually less than 900 Da). These molecules can be designed to interact with specific molecular targets in the body, such as proteins or enzymes, to achieve a therapeutic effect.

A small molecule drug is typically administered orally, as it can be easily absorbed by the body and distributed throughout the bloodstream. It can be used to treat a wide range of diseases and conditions, including cancer, infectious diseases, and metabolic disorders.

Small molecule drug examples include aspirin, acetaminophen, and statins used to treat cardiovascular diseases, while the targeted cancer therapy such as Imatinib and Sorafenib are other examples.

Harry Glorikian: But let’s take one step back. Right. So, you know, delving into this, it when I was looking at it. Right. If you just think about it objectively, it sort of seemed. counterintuitive and unreasonable since most of the successful algorithms in the world of deep learning are changed on trained on, you know, huge amounts of data. Right. And. By the way, those are the kind of machine learning we usually talk about here on this show, right, where the role of big data and medicine is is sort of one of our big, major themes. But but the results you got in the science paper were based on just 18 examples of known RNA structure. Right. At what point did you know that this was something that you could commercialize apply to drug discovery? Right. And the science paper only came out in like August of 2021, so it wasn’t that long ago. So how do you think about those two things of “small” data, right, in some sense, versus the large data. And how did you come to the epiphany that while we can actually do something with this and make money. 

Raphael Townshend: Right. I mean, I guess there’s two components to that. So the small data versus large data, right. You know, I think a lot of the key in the areas example specifically was that we had to very intelligently design the machine learning algorithm to be data efficient, specifically. So that it was able to sort of almost be aware of some of the laws of physics that drive molecular structure. In particular, you mentioned the rotation side of things, right? If you rotate your entire molecule, right, it’s still the same molecule at the end of the day. And more traditional machine learning algorithms essentially are unable to encode that directly. It kind of has to learn the thing separately in every different orientation. And so by building in this notion of rotational invariance, or really equivariance, which is a slightly related concept to all of that, we’re able to be much more efficient and work off of a much smaller data set. Right. Don’t get me wrong, additional data can almost always be helpful, but in this case, getting even a single RNA structure can take months. And so you don’t really have the luxury of collecting very large amounts of it unless you’re willing to spend a lot of time and money to really an absurd scale. 

Harry Glorikian: So, you know, hopefully I got this right. But if one of the big strengths of ARES was that it can predict whether a given molecule structure is accurate based on only a few examples of known structures. And it doesn’t matter what kinds of structures. In fact, if my understanding is correct, it doesn’t need to know anything about what types of molecule it’s modeling. It just needs to know where each atom is in a 3D space and what element each one represents. And first, I should ask you, did I even get that right? 

Raphael Townshend: Yes, that is is 100% correct. 

Harry Glorikian: Okay. 

Raphael Townshend: It’s reading 3D space and what kind of atom. Because at the end of the day, you could almost derive everything else from that. If you knew where all the atoms were, you could back out which ones were bonded together, know what elements, what bases those are comprised of, etc. And so that’s why a very similar system could be used on RNAi or on proteins or on anything else. 

Harry Glorikian: So that leads me to the question, why RNA and why not proteins? 

Raphael Townshend: I think that the protein space is very interesting as well in many ways. However, we specifically identified RNA as much more neglected from this standpoint, again. I think that’s a concept that comes back fairly often. The number of companies out there working on AI for proteins such as like designing antibodies or doing AI discovery of small molecule drugs to bind proteins—there’s a fair number of them at this point. And while that’s very exciting work, if you look at the number of companies doing AI for RNA structure, you know, it’s very limited. I mean you might argue that to the best of our knowledge, we’re the only one today. I don’t think that’s going to be the case for much longer, but it’s a very interesting position to be in. 

Harry Glorikian: Okay. Now let’s talk for a minute about what specific types of RNA you’re interested in, right? Because you know people who’ve heard about RNA and know what it is. I think there’s a misconception that there’s only that the only role, right, is as a messenger, right? Carrying genetic information from the DNA into the nucleus to the ribosomes where it gets translated into proteins. Right? This is standard. 

Raphael Townshend: The central dogma.  

Harry Glorikian: Right. But but there’s messenger RNA and then there’s non-coding RNA, right? And it turns out that there’s many roles for noncoding RNA in biology. I mean, could you explain some of these and say how non-coding RNA fits into healthy cellular functioning and disease? 

Raphael Townshend: Yeah. Overall, you know, there’s these interesting stats out there that you can look at. And just before I dive into the specific mechanisms, but if you look at the entire human genome, about 1.5 to 2% of it becomes translated to protein at some point or other, versus about 80% of the genome gets transcribed to RNA essentially. So there’s this vast space of RNA biology that’s still actually relatively poorly understood, and it represents really this opportunity to find that novel interesting biology. To speak about specific mechanisms, right, one very obvious one is you have the mRNA, which has this coding region in it, essentially, which is the part that codes for the proteins. But off the ends of these are these regions called the untranslated regions or UTRs for short, right, that essentially are not translated. The name is what it does essentially. And those pieces are oftentimes responsible for regulating various aspects of how quickly the RNA is degraded, how much protein is expressed off of a given transcript, or even things like determining where the RNA gets localized within the cell. Right. And so overall, there’s a number of these elements, these non-coding elements even attached to existing mRNAs that are responsible for regulating a lot of that function. Another example, right, is ribosomal RNA. The ribosome itself for translating from RNA to proteins is itself primarily composed of RNA. Then you even have, as you were referring to, all this non-coding RNA that isn’t even a part of RNA, which is even larger, vast world of RNAs whose functions sort of are still quite poorly understood in many ways, but have a number of roles that have been implicated in various cancers, for example. 

Harry Glorikian: So, you know, other than trying to ask you for, you know, what diseases might fit into this, right, from a non-coding RNA and where you could intervene is, why have these beenn and I think these are called undruggable diseases up to now. Like, why has it been so hard historically to predict RNA structures and find molecules that will bind to it? 

Raphael Townshend: Right. Yeah. So you mentioned this concept of undruggable diseases and a lot of ways people you know, this oftentimes refers to people trying to drug the proteins implicated in a disease and being unable to do so. You know, a classic example that people give of this is the C-myc protein which is involved in essentially overexpressed in 75% of human cancers. And we’ve been trying to drug this protein for a good 30, 40 years at this point and essentially have encountered a great amount of difficulty in doing so. And that’s essentially because the protein itself is mostly disordered, doesn’t adopt a single structure, and the binding interface when it does bind to other proteins and DNA, it’s in fact a transcription factor, is very large and so hard to interrupt with a small molecule drug. And so the dream of RNA is you’re instead going one step earlier in this whole process and instead targeting the RNA, the myc transcripts that code for this protein, and decreasing the amount there instead, which is really the objective that you’re going after, decreasing the amount of the C-myc protein in this case. And so that sort of maybe at a very high level is the dream of RNA targeting drug discovery or at least one of the mechanisms by which this happens. There’s actually a few different ones. 

Raphael Townshend: This dream, r a number of companies realize this and have been going after this this area for half a dozen years at this point, at the very least. But they’ve encountered a significant amount of difficulties. In particular, they found it’s hard to get a molecule that is selective to a specific RNA and doesn’t bind many other areas, which has obvious toxicity effects. But as well, really getting molecules that are functional, that can bind those RNAs and actually have the desired effect. And what’s emerged is that to get over these problems, structure really emerges as a very key piece of the entire puzzle, because if you’ve got a very unique 3D structure, first of all, that lets you very selectively target that structure and predict that. But then on top of that, if something is well structured like that, generally, you know, nature doesn’t waste effort. It’s something if it’s built something that intricate, that likely is functional as well and is doing something interesting. 

[musical interlude] 

Harry Glorikian: Let’s pause the conversation for a minute to talk about one small but important thing you can do, to help keep the podcast going. And that’s leave a rating and a review for the show on Apple Podcasts. 

All you have to do is open the Apple Podcasts app on your smartphone, search for The Harry Glorikian Show, and scroll down to the Ratings & Reviews section. Tap the stars to rate the show, and then tap the link that says Write a Review to leave your comments.  

It’ll only take a minute, but you’ll be doing a lot to help other listeners discover the show. 

And one more thing. If you like the interviews we do here on the show I know you’ll like my new book, The Future You: How Artificial Intelligence Can Help You Get Healthier, Stress Less, and Live Longer.  

It’s a friendly and accessible tour of all the ways today’s information technologies are helping us diagnose diseases faster, treat them more precisely, and create personalized diet and exercise programs to prevent them in the first place. 

The book is now available in print and ebook formats. Just go to Amazon or Barnes & Noble and search for The Future You by Harry Glorikian. 

And now, back to the show. 

[musical interlude] 

Harry Glorikian: So now that we’ve sort of covered all the science and I know it’s not all the science and some of the history background, let’s come back to the technology at Atomic right now. You’ve given it a new acronym. You’re calling it PARSE, which is Platform for AI driven RNA Structural Exploration. So first, can you explain, does PARSE include the areas and in what ways does PARSE go beyond ARES? 

Raphael Townshend: PARSE does include ARES, but in many ways the updated name is actually a reflection of the progress that we’ve made on this core platform. Since the seed round that we raised, we’ve actually been laser focused on improving this core structure engine, and that includes both sort of improvements on the AI side of things, the new algorithmic advances that even have happened in the last couple of years, even in the in-between time since the Science paper came out, as well as advances on the data side of things. We actually have in-house wet lab capabilities that have been generating data specifically tuned for the AI algorithm. And so part of the key of PARSE is that it’s no longer just a computational sort of algorithm, it’s really an integrated wet lab / computational technique with the two aspects really working in a tight, iterative loop to really accomplish dramatically improved performance and speeds in terms of the predictions. 

Harry Glorikian: So what kind of data are you getting from the lab, I guess the two pieces together, and how does it improve, I’m assuming it’s feeding back and it’s improving the prediction algorithm. 

Raphael Townshend: Yeah. The data in the lab are these genomics based assays primarily which are designed for scale, maximizing information content, etc. And how we leverage those is that we’re using them to supply and train the AI algorithms. And then once you’ve trained the AI algorithms, you can actually then further guide further experimentation in the other half of the loop and say, Hey, we really should collect additional data on these kinds of RNAs to really maximally inform the AI and see where it’s not confident, essentially, in this loop. On top of that, you can also use the experiments to validate some of the most key predictions. So if you were very confident you’ve got a structure somewhere, you can then go and validate that structure using more classical experimental methods. But overall, there’s really this sort of key integration of these two aspects of this loop of the experiments to train the AI, the AI that can guide for their experiments as well as be validated by those experiments. 

Harry Glorikian: So here’s the billion dollar question, no pun intended, after raising that much money. But ARES was cool because it’s so good at saying whether a predicted RNA structure was accurate. But now how do you transform that into a drug discovery machine? I mean, how does PARSE determine which parts of an RNA molecule are what, say, biologists call lgiandrrable or targetable with a small molecule drug? I mean, can you walk us through maybe an example or what you’re thinking? 

Raphael Townshend: The key overall is that while there are structured pieces of RNA, a large fraction of that RNA is also relatively unstructured and non functional or non ligandable either. And so really I describe it as you’re not only predicting those 3D structures, but you’re also finding which parts of the transcriptome, the space of all human RNA, is structured and ligandible in the first place. And this is really, you put it as billion dollar question, I actually think that’s very accurate in many ways. Because there’s this dream of targeting RNA. But if you talk to executives in the pharmaceutical industry, the difficulty is say you want to target the myc transcript, that cancer target I was just talking about before, you don’t even know where to start. You don’t know which part of the transcript to try and hit. And if you can hit it, if you can design a molecule that will be selective enough to hit it, or really accomplish the desired functions, and so really you need a technique to find these structured regions across the transcriptome, find these unique ligandible regions. And if you look at classical experimental technique, wet lab techniques such as X-ray crystallography or cryoelectron microscopy to name a couple, those are very good at validating a single transcript or a single piece of RNA if you’re pretty confident it’s structured. That’ll still take months or years if it’s doable at all. But they’re not really a tool for discovery. You can’t use it to scan through entire transcripts to find those structured pieces. And so that’s really where the AI plays a key role as you’re using the AI to scan through it and find those structured pieces in the first place. And then from there you can validate it through experimental techniques if you really want to. 

Harry Glorikian: So now that brings the question of what’s the business model for a startup like Atomic? I mean, your main asset is you have this proprietary machine learning model. Will you provide drug target leads to other companies as a service? You will you focus on discovering and developing your own drugs? You know, and if and if it’s the latter, is there a particular disease area you’re targeting? I mean, I was reading the press coverage and I’ver seen mentions of cancer and neurodegenerative, but I don’t want to make any assumptions without you telling me what direction it’s going in. 

Raphael Townshend: Of course, really how we’re looking at this is through a partnering sort of focused model. Right. We’re we’re revealing many, many structures across the transcriptome at this point, many more than you could ever possibly think of pursuing internally. You know, if you have a hundred structures and each one of those could potentially become its own drug program, it’s a little mind boggling at some point. And so by finding the right pharmaceutical partners to work with, we can really partner with them and build out these programs and partnership, really. And, you know, it’s not like a service kind of model. It’s really these co-discovery type of drug discovery partnerships that we’re looking at on that front. That being said, we also really want to focus on validating the structure internally as well and showing that you can use these structures to lead to promising drug discovery programs. So we’re also having our own internal efforts that we’re pushing forward to really validate the value of the structures we’re discovering. And so overall, it’s really this joint experiment, this joint internal plus partnering kind of model that we’re pursuing. And in terms of the specific areas that we’re looking at, we’re not disclosing at this time our specific targets. But there’s a couple areas where RNA-targeting small molecule drugs have shown a lot of promise. In particular, that’s oncology neurodegeneration, neuromuscular diseases, rare diseases and infectious diseases, actually, where there instead, you’re actually targeting viral genome. RNA, essentially RNA genomes. 

Harry Glorikian: So. I mean, I know we have a broad base of listeners and I just want to sort of maybe hit on a point, to make sure there’s no confusion. So I want to talk about how Atomic is different from say, other biotech companies focused on RNA, I mean, I think a lot of listeners are probably familiar with, you know, the kind of RNA targeting drug, namely antisense or RNA interference drugs developed by companies like Alnylam or Ionis. I mean, they make drugs where the business end usually consists of nucleotide sequences that are sort of complementary to the sequence known of some known RNA molecule. I mean, if you know the sequence of your target RNA molecule, you know in advance that your drug molecule will bind to it the same way that say, two helices of DNA bind to each other, and then you can sort of interrupt its function. But you’re talking about a totally different kind of binding that has nothing to do with the RNA’s nucleotide sequence, right? I mean, can you say a little bit about more about the difference? Why can’t you use antisense or RNA interference techniques to target the kind of RNAs that you want to target at Atomic? Or, I don’t know, maybe put differently, why isn’t it enough to know the RNA strand’s nucleotide sequence to know how to bind to it? 

Raphael Townshend: A lot of these ASO, SiRNA kind of technologies, we refer to them as RNA-based medicine, where the therapeutic is an RNA itself. Which is somewhat overlapping, but a distinct concept from RNA-targeting medicines where you’re really designing something to target those. In our case, we’re looking at small molecule RNA targeting, specifically. Designing small molecule drugs to go after RNA. And you know, these RNA-bases medicines such as ASOs from Ionis, they’ve done amazing things. But a major challenge with them is essentially delivery. Getting those to the right place in your body is quite difficult, essentially. You can target the liver. Everything ends up in your liver. You can target the eye, because that’s a relatively closed system. But for another, you know, for targets outside of that, it actually becomes a major challenge. You know, to give a quick example here, spinal muscular atrophy or SMA had an approved ASO from Ionis, Spinraza, in that case. But the delivery vector was essentially, you know, you essentially needed this surgery every month to deliver the ASO payload essentially. And then now instead, they’ve approved a small molecule drug to also target, very similarly, SMA. And in that case, right now it’s just a pill in a bottle that you just take orally. And the delivery system is just much simpler in some ways because we understand how to deliver small molecules in a way that these RNA based therapeutics are still, it’s very much an open research question. The difficulty, though, as you alluded to, is for these ASOs, you find the complementary sequence, you can target it that way versus to get a small molecule drugs targeting a given piece of RNA,  rthere’s no such thing as a complementary sequence anymore, right? There’s just a large, much larger space of chemistry that you need to explore. And so to target it productively, you really need to find these unique binding sites, right, these binding pockets, really is how we refer to them. And that involves really you need to model the structures really at the end of the day. 

Harry Glorikian: Interesting how we always seem to come back to structure being something that gives us a lot of insight about how to attack. 

Raphael Townshend: That’s how I view the world. At the very least, you know. M aybe it’s a slightly biased perspective. 

Harry Glorikian: But no, I mean, you look at AlphaFold and you think about I mean, there’s all these areas where we’re understanding how the thing is shaped and it really gives us insight about, okay, how am I going to attack this beast sort of idea, Right. But. So what’s next for Atomic? I mean. Do you already have a pipeline of drug candidates? You know, I don’t know. What sort of milestones are you thinking about for the company. And I guess if I said, hey, if we were talking a year from now, what do you hope would be different? 

Raphael Townshend: We are really at this point, at a fairly early stage in the drug discovery process. We’re just right now sort of developing our emerging pipeline of programs. We’ve really taken this data driven approach of scanning across the entire disease relevant transcriptome to find which RNA targets are ligandible. Actually, a very important point is that if there is a transcript that is of interest and we don’t find any structures in there, that’s also a very important piece of information because that tells you, you should not waste your effort trying to go after that one in particular. Save you a lot of pain and suffering down the line. So in terms of milestones, really what we want to do over the next coming years is start pushing forward these internal programs and really demonstrate the validity of the structures that we’re finding and connecting those to the downstream sort of functional impact that we’re very much after.  

Raphael Townshend: On the platform side of things, we’re also looking to continue and expand our capabilities. There’s many related sort of problems beyond just predicting a single structure that all have to do with machine learning on atoms and RNA generally. And so expanding those capabilities is also very key direction. You alluded to one earlier. It’s not just about predicting a structure, but it’s finding where something is structured across a transcript, which is a related but somewhat different problem. Another one is that RNA is a very flexible molecule. How do you model if it’s adopting multiple shapes? And that’s really linked to a lot of this exciting new sort of generative modeling   technology that’s been emerging on the AI side of things over the last couple of years. So there’s a sort of whole wealth of sort of problems to attack on machine learning on atoms in this RNA space that we’re really excited about, and that’s perhaps linked to the name Atomic AI at the end of the day. 

Harry Glorikian: Well, it’ll be it’ll be interesting to catch up, you know, a year from now and find out where you guys are and what you’re doing. I mean, I think, you know, I wouldn’t be doing this if I didn’t think it wasn’t super exciting and going to be very impactful to, you know, how do we treat certain diseases and, you know, help patients get better. So, you know, I can only wish you incredible luck. And it’s been great having you on the show. 

Raphael Townshend: Yeah. Thank you so much for having me. It’s really been a pleasure to be here. 

Harry Glorikian: Excellent. 

Harry Glorikian: That’s it for this week’s episode.  

You can find a full transcript of this episode as well as the full archive of episodes of The Harry Glorikian Show and MoneyBall Medicine at our website.  

Just go to and click on the tab Podcasts

I’d like to thank our listeners for boosting The Harry Glorikian Show into the top two and a half percent of global podcasts. 

To make sure you’ll never miss an episode, just open Apple Podcasts or your favorite podcast player and hit follow or subscribe.  

And don’t forget to leave us a rating and review on Apple Podcasts.  

We always love to hear from listeners on Twitter, where you can find me at hglorikian. 

Thanks for listening, stay healthy, and be sure to tune in two weeks from now for our next interview. 

FAQs about Small Molecule drugs and more

What are biologics vs small molecules?

Biologics and small molecules are two different types of drugs used to treat diseases.

Small molecules are chemically synthesized compounds with a low molecular weight that can be orally administered and easily absorbed by the body. They can interact with specific molecular targets in the body, such as enzymes or receptors, to achieve a therapeutic effect.

On the other hand, biologics are large, complex molecules that are derived from living organisms, such as bacteria or mammalian cells. They are usually administered via injection and work by targeting specific proteins, cells, or other biological molecules in the body.

Biologics are generally more expensive to produce and administer than small molecules due to their complex manufacturing process. However, they can be more effective in treating certain diseases, such as autoimmune disorders, because they can target specific immune system components with high precision.

Examples of biologics include monoclonal antibodies, vaccines, and gene therapies, while examples of small molecules include aspirin, ibuprofen, and statins used to treat cardiovascular diseases.

how important is small molecule manufacturing?

Small molecule drug manufacturing is essential for the production of a wide range of drugs used to treat diseases. Small molecules represent the majority of pharmaceuticals on the market and are the most commonly used drugs for the treatment of various medical conditions. They are also widely used in research and development, both in academia and the pharmaceutical industry.

The manufacturing of small molecules is critical to ensure the availability of safe and effective medicines for patients. The process of small molecule drug manufacturing involves the synthesis, purification, and formulation of the drug substance, as well as quality control and testing to ensure that the final product is of high quality, efficacy, and safety.

Efficient and cost-effective small molecule manufacturing is important for ensuring the availability of affordable medicines to patients worldwide. It also plays a critical role in drug discovery and development, where efficient synthesis and purification of small molecules are required for identifying and optimizing drug candidates.

Overall, small molecule drug manufacturing is a crucial component of the pharmaceutical industry and is essential for the development and production of a broad range of medicines that improve human health and save lives.

how does a small molecule drug work?

Small molecule drugs work by interacting with specific molecular targets in the body, such as enzymes, receptors, or transporters. They are designed to bind to these targets and modulate their activity, leading to a therapeutic effect.

When a small molecule drug is administered, it is absorbed into the bloodstream and distributed throughout the body. The drug molecule then interacts with its target, which can be located on the surface of a cell or within the cell. The binding of the drug to the target can have various effects, including:

  1. Activation of the target, leading to an increase in its activity
  2. Inhibition of the target, leading to a decrease in its activity
  3. Modulation of the target, leading to a change in its activity

The specific effect of the drug depends on the nature of the molecular target and the mechanism of action of the drug. For example, an anti-inflammatory small molecule drug such as aspirin inhibits the activity of an enzyme called cyclooxygenase, which is involved in the production of inflammatory molecules called prostaglandins.

In summary, small molecule drugs work by selectively interacting with specific molecular targets in the body to produce a therapeutic effect. The efficacy of the drug depends on its ability to bind to the target with high affinity and specificity, and the safety profile of the drug depends on its selectivity and lack of off-target effects.

What is RNA in small molecule drugs?

RNA (Ribonucleic Acid) is a molecule that plays a critical role in the genetic expression and regulation of cells. It is a nucleic acid, just like DNA, and is made up of a long chain of nucleotides, which are the building blocks of nucleic acids.

The nucleotides that make up RNA contain a sugar molecule, a phosphate group, and a nitrogenous base. The nitrogenous bases in RNA include adenine (A), guanine (G), cytosine (C), and uracil (U). These bases pair with each other to form complementary base pairs, which are important for the structure and function of RNA.

RNA is synthesized from DNA through a process called transcription, where a portion of the DNA sequence is used as a template to create a complementary RNA sequence. RNA can exist in different forms, including messenger RNA (mRNA), transfer RNA (tRNA), and ribosomal RNA (rRNA).

mRNA carries the genetic information from DNA to the ribosomes, where it is translated into proteins. tRNA is responsible for bringing the correct amino acids to the ribosome during protein synthesis. rRNA is a key component of the ribosome, which is the molecular machine that synthesizes proteins.

Overall, RNA plays a critical role in the regulation of gene expression and protein synthesis, making it essential for the functioning of cells and organisms.