Lantern Pharma Inc. ($LTRN)

Well, first of all, I want to thank everyone for joining us at 8:30 Eastern to see what I think is going to be probably the first real public unveiling of this next-generation withZeta AI platform. Many of you have had some of the fortune to actually see it in the past in demos. Some of you are actually users, which is even better. But we've now come up with the next generation. And let me walk you through some of the features. Just as a reminder, withZeta.ai started as an initiative at Lantern Pharma for us to think about rare cancers. And we've been particularly, I would say, both gifted and focused on trying to take our therapies and focus them on challenging or rare cancers, partly as part of a development strategy, but partly also their white space. There are a lot of cancers that basically have no standard of care or a highly unmet need -- have high unmet needs. Zeta is actually one of the rare stars. It's a type of star, Zeta star, and they're very rare. And so as we kind of thought about this project, we codenamed it withZeta, initially Zeta and then withZeta because as the power of the platform increased, it became more than just a big data platform. It became more than just a RADR platform. It became more than just a platform for going out and gathering information and putting it into nice tables. It does all that. But it actually started having the ability to reason. We use natural language processing and we tied all our tools together. And it actually became almost like a colleague, the co-scientist. And so we said this is really the withZeta project. And we focused on rare cancers. And there's over 438 rare cancer areas that we'd like to conquer. And they're going to be accessible to any investigator, any scientist anywhere in the world. And that is very, very important. So the market around rare cancers has largely been not very successful. It's usually done as an afterthought of other drugs. You'll have a drug, you'll have a molecule and then maybe you're fortunate enough to think about the mechanism, perhaps working in a rare cancer. And what makes rare cancer challenging is that the markets are tiny. But the deaths in aggregate are almost 30%, and there's about 5 million rare cancer patients globally, very often without any therapies. And oftentimes, that trials, oftentimes that enough data for someone to convincingly go after something, and what we did is we accumulated all that sparse data, all that information, all the disease models, all the approved drugs, all the clinical trials that have failed, all the clinical trials that have succeeded, all the literature that really makes sense and put that all together in one place to help us reason through all this. So as the tool became more powerful, largely because of Reed and his team doing the data engineering, we're able to develop a tool that became a platform that became a co-scientist. And it really works with us, and we'll talk a little bit later about what it's done not only for Lantern, but for others. And on top of that, we're able to not just repackage information, but generate net new scientific knowledge, and we'll see this today. We can bring in real-time literature search, do molecular pathway analysis, do all kinds of molecular analysis around a molecule, actually generate new molecules using a 24 billion parameter LLM focused just on molecular structures that are medicinally appropriate. And then actually also live and have all that knowledge as a knowledge graph. And so -- and it builds on that knowledge graph, and we'll see that. We designed it so oftentimes, people want really, really in-depth information. So one of my colleagues, Chief Scientific Officer, he may not want 1 or 2 paragraphs. We may want 6 pages. Maybe there's others that want 19 pages. Some people may want to report. Some people may want to have quick answers that allow them to then have more investigations later. And so we created research modes, much like you'll see in many GPTs. You'll have a quick explorer mode, which is actually very efficient and there's multiple rounds. So it's really a scientist. It's really not, hey, how's the weather? It's really still thinking through some complex initial exploration. An investigator mode that recurses through 10 investigation rounds with lots of parallel tools. And to put that into context, imagine if you're -- any of you who are chess players or game players, thinking through like 10 steps ahead and oftentimes 3, 4, 5, 6 games at a time and 10 steps ahead, it becomes quite complicated. I know I can't do it. Maybe there are some super humans that can do that kind of stuff. But imagine now being able to do that in some complex disease questions in cancer. It's really an amazing thing to see. And it's not hours. This is literally minutes. And then finally, you can put all this together and you can share the knowledge graph with others, but you can also share the report. And on top of that, what we've done is thought about what are the different types of tools, what are the different kinds of people that it takes to build great medicine. And it takes a lot. Not only do you need general research scientists, but you need people who are highly trained at molecular feature analysis and scaffold design and ad lead predictions. You need the medicinal chemists. You need to have people who think about the trials. How do you develop a protocol? What's working, what's not working? What are things that are really endpoint strategies? Or what's appropriate from a regulatory standpoint? Then you also need the concept of a real person in the clinic, an oncologist who understands the landscape, knows what standard of care, when to deviate, when biomarkers are useful, when they're overkill, and really what is the strategy to go from a molecular concept to a clinical translation. And clinical oncologists are so wonderfully gifted at that, and they're seeing patients. And then finally, the translational scientists. What is the biomarker validation that we need? What's the correlations that make sense? What's already been done? What's worked, what hasn't worked? And so all these co-scientists are there at your disposal. One of the things we try to do also on top of that is we built proprietary knowledge bases, things that we curated that we as Lantern scientists and Lantern developers think that this is the way we want to make drugs. This is what we would expect from our colleague. And so we've kept the framework simple, bulleted, detailed enough and encompassing always pushing forward to get to patients. And so we'll talk -- we'll take a look at the demo today, but now you can actually tune Zeta to take on the personalities of these various co-scientists. And maybe, Reed, you can expound a little bit about how we did that and then where that's going next?

Yes. So for those of you all in attendance that have been fortunate enough and we're grateful that you've been with us through the beta experiment, you've been discussing with the general research scientists. And a lot of the feedback that we got was contradictory in some ways, where some users wanted super in-depth clinical trial protocol design, while others wanted just quick, generate a molecule and let's iterate on the molecule itself. And so it wasn't that one of those responses or feedbacks was more accurate than the other, but it really highlighted the need that there are these different co-scientist personas that personify or embody the type of scientists that you want to be discussing with. So the way that we've done this is, all of these co-scientist personas have access to the same underlying databases and tool sets and skills. What we've done is tune which tools that they are preferential towards and how they then organize those tools into sequential and coherent workflows to actually get to answers. And so the medicinal chemist is going to focus on actual scaffold based or de novo molecular generation with very rigorous chemical -- actually computing the molecular features of those generated compounds to validate, whereas the clinical trial strategists and clinical oncologists are going to rely far more heavily on AACT and our clinical trials database and our rare cancer knowledge base. So the same information is available to all of them, but it tunes how specific they are to various domains.

And this is really what happens in the real world. So if you're a knowledge worker and you have a lot of things, you are multidisciplinary, everyone is going after the same problem, but the tool set that you bring and the knowledge base that you bring oftentimes is slightly different. Think of a house, the person in charge of plumbing obviously knows how a house works, but they're going to focus on the plumbing set and the tools. The electrician, same thing, they're going to focus on their tool set. And so in any multidisciplinary knowledge world, they all are focused on the same problem, but the tools they bring out and the tools they prioritize are slightly different. And that gives you some really interesting different responses, especially as the responses cascade over multiple iterations. And we'll talk about, this is very theoretical, but what we're thinking about in the future with these different scientists that are powered this way, is to actually have the scientists talk to each other. And maybe, Reed, you can expand upon this concept that we're beginning to play around within build.

Yes. So now that we've expanded the platform to embody these various co-scientist personas, the next step for us is, two kind of scopes. The first one is that if you're talking to the general research scientist, there might be a question where the general research scientist wants to go off and consult the medicinal chemist. So in the general workflow without any intervention by the user, Zeta can kind of route the question towards the appropriate expert, essentially at a higher level of abstraction, but emulating the mixture of experts model in LLM engineering. But then beyond that, what we're considering is this idea of the simulacrum, which by a strict definition, it means that the simulation or emulation of something becomes so real that it replaces or appears just as real as the original thing that it was set to imitate. And our intention with this is to create kind of a roundtable conversation of panelists where these various co-scientist experts are talking to each other with a high-level prompt to kind of seed question of what you'd like to discover or ask about. And then the human is still there in the loop. The human is one member at that round table. But now instead of you with a single co-scientist, it's you with a collective of intelligences.

Yes. And this is very important because we really think this eventually will allow companies and teams to have an entire biopharma development team in a box. So imagine you as a Chief Scientific Officer or as a Lead Research Scientist or even as a CEO, and your job is to create some unique novel therapy or to find something interesting, not only can you have one team of scientists at Zeta, our eventual vision is you can have multiple teams going after these problems. And we'll talk a little bit about that later on. But let's get down into some of the things that we're going to talk about today in the demo and what you're going to see real time. We're going to give you -- Reed and I are going to provide you an orientation to the broad platform, features, things, buttons, just how do we use it? We're going to do some queries on some rare cancers to show how the investigation works. We're going to see how the knowledge graph can be used and how it's a great feature, especially for teams and organizations. We're going to do some real-time molecular analysis, including some really detailed predictions and then actually maybe think about generating a novel molecule, either based on a de novo structure that we uncovered today or a scaffold that we think is promising or some new target that we think needs to be interrogated. And now we're open to that. Now remember, I want to remind all of you that this is not canned. This is a live demo. Now some of the questions, of course, Reed and I did go over in advance, but this is live. You're going to see the system live, and we've engineered it also with a lot of cost in mind. So everything that you're seeing is being terraformed real time, and that's very important. This infrastructure-as-code is also what allows us in many ways to have some controls around performance and cost. So let's go through for our attendees now to the live and get rid of the PowerPoint, and move over to the system. [Presentation] Okay. So this is where you would land. So this is where -- if I've set up an account, I would go here. And I could pick the persona or keep it general, and then I could ask a certain question. So is there a question anyone wants to type that they want to ask today about a rare cancer, anything specific?

Not sure if we have a question box for the attendees or any way to pull that. I would say just run with it, Panna.

No problem. So let's ask it about over-expression of EGFRvIII. What rare cancers over-express EGFRvIII and still are in need of improvement in therapy? So let's ask you that question. I'd like to go into investigator mode. And as you can see here, we're doing the general scientists. I could pick any specific scientist. But we're going to do general initially, investigator. So these are some of the things that you may want to think about as you type your question, like what mode, et cetera. And of course, you want to be cognizant about your credits. So you'll have your credits here. And obviously, the more depth, the more credits that will get used. But let's go ahead and ask this first question. And EGFRvIII is a topic. It's definitely a target. It's definitely a concept that cancer scientists think about. And so now it's going to start thinking. And unlike a lot of AIs, what we're trying to do is walk you through the reasoning process and what tools are being used and how it's investigating it. So just imagine if you had someone in your team and you say, "Hey, give me what's going on in EGFRvIII in these hundreds of rare cancers because we want to prioritize it because we have a molecule that we think really does something in that space. Great. And then more importantly, it's going to read all of these papers or review these papers real-time. This is something that would take hours, maybe days, unless you happen to have a colleague that already knew everything about EGFRvIII in rare cancers. But now it's going through and it's giving you kind of real-time what it's thinking about, what it's researching and reviewing. And then it's going to use tools to execute across that in terms of searching, in terms of reasoning, and then it's also going to give you the reasoning process. And this is very important because the reasoning process, if you're a company, let's say, you're Amgen or Servier or BMS, you may want to tweak it. You may want to say, "Hey, this is the way we like to think about things. We'd like to put this higher, this lower. We'd like to add these concepts or we'd like to give it a certain flavor and index these internal data sources first". And these are all things that can be quite readily tweaked for the enterprise features, and we'll talk a little bit about that as well. So now based on the initial findings, it's going to give us, again, an answer.

Panna, by the way, if you want to collapse those sources, you can click the source on the top right there.

Yes. So we can see it's gone through about, what, 30...

41 in this case. So those are all the external resources that it's querying. That's going to be the list of clinical trials that it's referenced, PubMed articles that it's found the abstract for and DOIs. And then additionally, if you wanted to see other internal proprietary databases and AI modules that were queried you could go into the research queries and see exactly what tools were executed in this process.

And so now you have a fairly detailed answer from Zeta that says here are the therapeutic gaps, primarily affected GBM in rare cancers, giant cell GBM and gliosarcoma, similar GFR fusion patterns, overlapping mutation landscape. And what you can notice is the colors, the colors are very important because as drug developers and people in science, we'd like to have things that pop out as we read thousands and thousands of words, right? And so things that are disease are in red, things that are biomarkers are in green, literature references are reversed in the gray and white, drugs are in orange. And so quickly, as you see these colors, as you scan the landscape, you can see, this is a drug and we look at that, and we dig into a disease. And we look at a trial that's being referenced. This is a [indiscernible] vaccine, which is spectacular Phase III failure, okay, immune checkpoint. So it's going through the litany of drugs being used, the mechanisms driving some of the resistance that it sees. So this is, again, thinking like a drug developer. It's not just giving you a summary, it's thinking about what can I do to improve the landscape. And part of that is always looking at what is resistance, what's not working, lysosomal sequestration and emerging strategies to overcome resistance, multi-antigen targeting, combination immunotherapy, TME modulation and CD47. And then tells you what the unmet need is, young patients, bad prognosis. So it gives you a pretty good overview. And again, it gives you insight into that space. Now what it did as it did all this, which is very important, is it generates a knowledge graph. So much like any knowledge worker, you think about these concepts and how they're related to it. Sometimes there's concepts that float around that aren't perfectly related. But you can see EGFRvIII is over-expressed in GBM. GBM is treated with these drugs in the orange. So orange is drugs. There are certain proteins like LAG-3 or SERP or TIM-3 that may be involved. So again, you start creating the knowledge graph, and this may then get linked to other concepts like this disease, giant cell glioblastoma or gliosarcoma. And so this landscape continues evolving. So let's ask it about -- one of the new features that some of you that have been using Zeta in the past will see, we'll demo this next Reed is, we now have given Zeta, much like a lot of the AI tools, the ability to actually prompt you with questions, ways to clarify or improve the question. So sometimes if there's something that it wants clarity on or it says there's an overwhelming abundance of directionality in terms of the tree or the -- it will come and say, do you mean A, B, C, D or can you clarify? And that, I think, is also a great improvement because that's how people interact, right? You ask a colleague question like, Mike, hey, Mike, what do you think about this target, it was like, well, do you mean about it in the context of the intra-tumoral environment? Or are you talking about it as a therapeutic target? Those are the questions that any kind of co-scientists, your colleagues would have. And so now Zeta does that as well today. So now it's telling you it's looking at the late-stage landscape. No gliosarcoma-specific trials, although it does find some current standard of care. So this is great, and it has some evidence. This is going way back to the SCUBE trials.

Those are glioblastoma trials.

Yes. And then this is [indiscernible] it's including it, recent failed approaches. That's great. So let's say -- let's look at one of these in detail. In fact, this is a great place maybe for Zeta to say, I'm missing something, right?

Also probably a good chance to switch the co-scientists you're talking to.

Yes. Are there any deeper questions I should be asking or that I am missing? And let's switch to -- should we switch over to a biomarker and translational scientist. Let's try that.

That is an interesting thing. You can switch between depth and the co-scientists you're discussing with, middle of a conversation, seamlessly. So you don't have to start a new conversation to change the co-scientist persona that you're discussing with, kind of the hat that Zeta is wearing in that moment. So it makes it pretty easy to evolve a conversation thread in the way that we're doing here.

And again, as you can see, the knowledge graph continues to evolve. And so if I want to export the knowledge graph as an HTML file, as a JSON file automatically in store, which is very important because then it's machine readable for any future use. So now we ask it to -- there are certain deeper missing questions and we gave it the persona of the translational biomarker scientist. And we'll see what it comes back with. And then after that, what we'll do is we'll dig right into actually designing -- potentially designing a molecule or designing a trial and actually giving us a budget and time line and a protocol. So any of those venues -- and the important thing about the way we provided the tools and capabilities of Zeta is that we thought about it in the landscape of what do we as drug developers need to do. We need to put together disparate results. We need to think about scaffolds. We need to optimize molecules. We need to look at how molecules are made. We would love to do de novo design, and we do that also now. We'd love to maybe benchmark that. Future versions will include integrated bio-informatic tools, which obviously people will use and pay for and also be able to do it on any of the data that we've already pulled or even your own data. So now there's a mode you can actually upload information to Zeta as well, and we'll talk a little bit about that feature after this question.

Currently limited to PDFs and text files, but this will quickly be expanded to actual source data that you might want to have uploaded to enhance some of these questions for your own sake.

And you'll be able to do it eventually into your own secure Amazon or Google Cloud as well. So let's imagine you have a folder full of data on pancreatic neuroendocrine tumors. You want to say, "Hey, Zeta, I want you to go ingest this proprietary data. It will just be between you and Zeta. It won't train the rest of the -- machine won't learn on it. It will be your own private instance. And so those are things and features that we're building out now for future enterprises as well. So as you can see, this is great, deeper strategic questions you should be asking, therapeutic resistance questions, translational strategy, which is key mechanistic biology questions, clinical trial questions. And you can see here, it's used over 14 tools. So a lot of tool usage in this one to kind of give us these types of questions. And again, no dedicated drug development, lump with GBM despite being -- having distinct biology, worst treatment response, unique molecular features, no biomarker-directed trials, wow. So here's a clear gap, right? Prognosis reality, real-world outcome data, we can go look at it, terrible MOS and it gives us outcome in some of the trials, EGFR TKIs, ADC, that's an interesting ADC. That's pretty late-stage. Again, gliosarcoma is very, very challenging cancer. So let's look at this drug here, Afatinib. It's an EGFR TKI, which we -- a lot of people know about. And we'll ask Zeta a really important question, can we create an improved version of Afatinib that has combination potential with a checkpoint?

Might want to switch to the medicinal chemist for this mode, and that brings up the valid point what I mentioned before, if we're working on having Zeta actually do the routing across the various co-scientists appropriately on its own.

Now the future of this isn't just your one-on-one interaction with Zeta, I'm looking at this question, but also now you've created kind of a unique knowledge pathway that you can share internally in the organization. You've gotten tons of sources. So you have your own, what is this now, 90 some...

97, it looks like, yes.

Yes, that it's now looking at and ingesting. And more importantly, you've got a huge knowledge graph that's growing. And so all this can be shared and you can come back to it as well. So Zeta thinks it's an excellent question. Good. I'm glad. This is precisely the rational drug design challenge. I'm glad that's why we gave it to you Zeta.

We like to think so.

So now if I had to ask this question, let's say, a CDMO or my medicinal chemist or team, they'll go off for weeks or maybe months and think about it, think about the scaffold, what works, what doesn't work, what are the features that they would want to improve? What's going to make it work best with the checkpoint? How do the checkpoint inhibitors work? Which one features in the checkpoint would I want to synergize with so that I can have the kind of synergy scores and values that combination regimens have. Big complex questions, and we're giving it like buckets of different kinds of questions that can take drug developers months to even come up with answers that they feel are rational, grounded and have the ability to get validated in the lab. And that is very important because you're going to ultimately ask the question, well, can I take these concepts to the lab? And we check for that. So anything -- any molecule that comes up, it checks, can it actually make it and you design around a translational biomarker strategy. It things like a scientist and says, is there a proxy or are there methods? Is there a biology? Is there a publication that, hey, this is something that I can go take off and do through proteomic work, tissue microarray work, standard sequencing work. And so again, validation is very important. So it has the knowledge or awareness that the things that I'm going to provide to you as a user need to eventually be put and tested into the real world. So as you can see, it's beginning to use lots of tools, validating SMILES strings, computing molecular features, and it's going to do that, again, real time, its doing this in minutes, not days. And after it does all that, it also will update the knowledge graph and update the bibliography and so you can share that. As we look at that, we'll go through a couple of other features. So you can start your conversation. So if it's a conversation you want to have, you can see EGFRvIII therapeutic resistance. I've started that. And so that's kind of top of mind. All your recent conversations will be here, as well and then your research toolkit and then kind of what rare cancers. And so we have a rare cancer tree, oncology tree as well that's built into the tool. So you can see how all the rare cancers are grouped, classified and what families they belong to. Quick start guide as well, which we urge you guys to take a look at and also certain account features. So while we wait for it, we can go to the account features. And so the account features, you can see how you've been using it, how your credits break down, what conversations, persona et cetera. So all this is all available. So let's go back to our chat.

That's a different one.

There we go. All right. So did it create the molecule here?

I think because you went to the account page, it stopped the stream.

No. Okay.

So just ask the question again. Yes, I should have warned you that the toolkit, rare cancers, all of the other modals pop-up and will keep the conversation stream running. The account page will kill the web socket, unfortunately.

Well we're learning real-time. It makes sense the account page would kill the web socket, yes. Makes sense. Can you probably want to do a -- do you want to go to account page?

There's a way to make the account page, not break the web socket, future version update. So yes, while this is going, unfortunately, for the second time, one thing to note as well is the way in which this medicinal chemist is going to go about this. So as a scientist, you are not going to one-shot a brand-new molecule, whether it's de novo-based or scaffold based. It's going to take multiple iterations of trying a modification, recomputing its molecular features and seeing what dials got turned. Once you know that, then iterate a second time. So Zeta is going to do both of those. First of all, it's going to -- we have several tools, one in particular, which is computing the molecular features of it. And this is a subcomponent of our PredictBBB application, but it computes over 90 physiochemical properties of the chemical itself, such as molecular weight, hydrogen bonds, rotatable bonds, total polar surface area, a whole bunch of them as well as drug likeness scores and filters and then use those as guardrails for prompting Ether Zero, which is the smaller open source large language model, which is actually doing the SMILES modifications and reasoning in English. And so it's this back and forth from real-time computation of molecular features, validating how a compound actually behaves and what it looks like physically to then modifying it with Ether Zero and then repeating the loop until you've arrived at optimal features and targetability.

And as you can see now, it's up to 8 tools that's going through. And it obviously has it marching orders to try to use the scaffold around this targeted TKI, but improve the BBB of it and more importantly, also improve its potential to be synergistic with checkpoint inhibitors. So kind of given us some concepts, and it's going to come back and tell us if that's doable, not doable. And the thing about the co-scientist personality also here is that it likes to tell you when it can't do something. So unlike a lot of LLMs or chat agents that try to force an answer to please you, Zeta is not in that category. If they can't do something or if the drug is good enough or it doesn't seem like there's something doable, it will come back and tell you exactly that, which is great because it's more efficient that way and now you can go through a different strategy. As it streams the answer, you can see how it's thinking about it, is actually decent, real barriers are -- it's molecular weight too high, LogP too high, excessive flexibility, the rotatable bonds. So it's adding methoxy group to increase total polar surface area compared to completely wrong, then it says it wrong. New optimization strategy, keep the halogens, reduce [indiscernible] dimethyl, preserve the core, change the warhead. So now it's going to actually look at all of these very, very specific things. And it came on the optimization strategy by looking at all the molecular features, as Reed talked about. So the molecular feature tool that we have that PredictBBB, but also predicts tons of other stuff is looking at that analysis, feeding it to the other LLM. The LLM is then incorporating that for potential de novo design or scaffold improvements. And then it's going to then continue iterating also because we want the synergy with checkpoint. And just in terms of time, I want to make sure that we hit on a few other things. So it's going to continue working on this in the background. It's going to rephrase with some stronger therapeutic context, which is great. It's now up to 16 tools.

Yes, it's still going.

So as it continues to move, which is really fantastic, why don't we go over to some of the future things that we're going to do? I'm going to leave this in the background, Reed.

Yes. Just go to the PowerPoint.

Yes, I go to the PDF. So I'm going to do that. And then I'll keep an eye on when it's got done. It's making -- it's now -- it's through a second iteration actually it says. So it may be done quickly. So again, now we're generating a novel molecule, kind of the end of some of our real-time discovery. But what's next? So as we mentioned, enterprises, we're going to have features to have team workspaces, role-based access and audit trails, custom integrations with their internal R&D platforms and then white label deployment for big partners because they'll have their own cloud, they'll have their own data resources. And of course, they don't want to share it in the public with Zeta. We're going to have social features to have shared research environments for teams, credential research profiles, multiuser investigation sessions and then perhaps even feeds that are personalized that tells you what new data and information is available. We continue to enhance our tools, broader data modality support, potential IP landscape analysis and then increased integration with some of our proprietary tools like our pathway mechanism knowledge base, expanding the literature around the -- access around the biology models, specifically our ADC squared model, more disease oncology around rare cancers and then synthetic lethality and combination therapy prediction, which is a very important category. And of course, key things, optimized for mobile, structured on-boarding so that people can understand all the tools and capabilities and perhaps even in the future, regulatory tools supporting IND prep, therapy designation framing and actual document production. So these are all the things that Zeta will do. Let's go back and see how Zeta is doing in the molecular generation. Let me stop that and go back. Great. Well, that's good. So in the time it took to make a couple of -- show a couple of slides, it's created an optimized analog with enhanced BBB penetration and checkpoint inhibitor. So as you can see, it benchmarks it against the original improvement, what its targets were and did it achieve the targets. It's at the warning threshold for rotatable bonds, which Afatinib already was anyway because, in fact, that's one of the challenges with it, but it's improved the drug likeness. It passed 4 out of the 6 drug likeness filters that we've got. It made some modifications. It tells you the modifications just like a scientist would tell you. It tells you why it's going to be synergistic with checkpoint, which is great. This is really key, what the mechanism and activation pathways are. And then very importantly, it will give you some combination ideas. Dosing and sequencing strategy. So again, we've really designed this to, not just give you a one-off answer, but then push the answer into getting it to patients and trials. And again, it may be 100% right, it may be only 70% right. So if you have multiple people that look at this and you share the knowledge graph, a few hours of work with Zeta in a couple of days, you can do what traditionally has taken months. And more importantly, you're not going to miss key ideas. And it will tell you how a biomarker-driven patient strategy. So I can take that patient strategy. Imagine taking this, cutting and pasting it and putting it into the biomarker mode and saying what are some disadvantages of this approach or advantages of this approach. And I can also now take all of this and create a PDF of this entire thing. And so that's also very cool. So I can take this and just save it as a PDF. And so it will save it as a PDF. And as you can see, it's also now updated the knowledge graph also. So the PDF just downloaded. Let me bring it up, which is the PDF just downloaded onto my desktop somewhere. And since I have a very organized desktop, it will be very easy to find.

I'm not sure if we're seeing the window.

No, I got to switch over to the window.

Yes. While Panna is pulling that up, one thing I just add to the mention of this is that Ether Zero has now proposed this scaffold-based modification of -- for a new molecule. And then there are many different ways that we could take this, one of which being use Ether Zero again for actual retro-synthesis analysis and propose within the medicinal chemist a way to actually synthesize the molecule. Otherwise, as Panna said, we can go into biomarker strategies or create a whole clinical trial protocol around it. And again, that's where you can switch between the various co-scientists for the type of response you're hoping to get.

And then it will feed out a nicely laid out PDF with Zeta. It will have its appropriate disclosures. You know that this is really not a medical advice. It's not diagnostic information. It's really for specialized research and its information. So it tells you what its goal was, what its approach was really just a lot of the things that we covered. And again, keep preserving some of the highlights that you can tie all that back to the knowledge graph that you've got. And it's got the strategic recommendations as well and dosing and sequencing strategy. And again, every one of those areas can be improved. So you can go back and then test that. And again, our vision is that you'll have maybe an initial response with this kind of depth of 9, 15, 20 pages. And then you feed that to the entire community of scientists and you have them discuss and talk about and come back to you. So imagine if you're a team working on the future of metabolic inhibitors or metabolic pathway drugs, you can have multiple teams, all working like literally within hours and days to go attack problems that traditionally have taken scientists years. And our vision is what's taken 50 years in the past in cancer research and cancer drug development. Well, now in the future in the world of these multi-agentic infrastructure only take a few years. And this is really the CEO of Anthropic pushed out this concept, but being able to do in 50 years -- be able to do in 5 years, what used to take 50 years. And that's what's possible with these tools. And so with that, I'd like to thank you guys all for joining this morning. And we'll maybe open it up to any questions that people have or anything that we've got a few minutes remaining that people want to discuss, but I urge you guys to take a look at withZeta and withZeta.ai. We have a couple of tiers of subscription available. And for those of you that on the demo and knowledge accounts, it's great, keep using it. We'll have obviously the new version in production now. Is that right, Reed?

Correct. It is fully live in production with all of the new -- the latest and greatest you saw today as well as full payment integration. So if you were a user in our beta tier for early access, your credits and all of your access has been retained. You don't lose access now that we have launched to live. There is a grace period of about 2-months that you still have full access to the platform. And I'm not sure if we have it set up to receive -- we do have Q&A. Okay. Great.

Yes. I want to take some of these questions and Reed, I'll let you know. So we have an anonymous attendee. AI tokens can be very expensive, how profitable are they? The subscription plans are very, very, very profitable. Again, as I mentioned early on, we didn't dig into it. We really optimized around the token use, and we're -- we've got a lot of room we can continue improving it, but it's a lot of it has to do with credits and knowing what tools use, which credits. Also, a lot of our work isn't in inference. It's really in large quantitative models, LQMs. So all those molecular feature analysis that is the underpinning of molecular design, a lot of that is not done on inference chips like NVIDIA or Grok, those are done on traditional chipsets that are actually much faster and cheaper. So things like RAG and response and a lot of that uses inference chips, but a lot of the hardcore chemistry and some of the biological work where it's data-driven doesn't need the same kind of tokenization and is fairly inexpensive. But good answer. But yes, the subscription plans, again, are also guard-railed through credits. So that's been answered. Next question, does it access information and data on devices? Good question. I think in the future, we could, but the current version does not access information on devices. One could upload it. And Reed, do you want to talk a little bit about the upload feature?

Yes. We do have -- this is again, the latest version that is now on production withZeta.ai, but you can upload text documents. So document files, text files or PDFs can be uploaded and included in the thread of a conversation. So it's not added to the global Zeta knowledge base or like a personal knowledge base of your own, yet. But what it does is it just adds the context or the content of whatever text file you've uploaded into the current context of that conversation thread. So that's how we've handled this at the very starting point for right now of just allowing users to upload PDFs if they want to upload an academic literature and ask questions and cross-reference against it. But that's currently the extent to which Zeta has any access to on-device information.

Good question. But in the future, you can imagine a big drug development team may have a sequencer or a proteomic installation or something where they're generating data on the daily, that can all be stored in their cloud. And then eventually, Zeta would have access to that in a real-time way just for that company or for that team. But yes, right now, we've guard-railed it from going in the broader system because we don't want the chance of actually any information not being used properly for any of our users, but more importantly, people wanting to try to poison the system, too. So the Zeta knowledge base is very, very closed. But good question. Any more questions? If not...

Yes. Panna, there was one more. Do you mind disclosing current subscription revenue and prediction of it in your next quarter? Disclosing subscription revenue.

We're just launched, so probably too early, but we do have some AI revenue already. So that's exciting. But we'll talk about the prediction and the growth, I think as the AI platform fully launches. We have a major launch coming up at the American Association of Cancer Research in San Diego later this month, and we've got a lot of demos already set up there. So I think this will probably do very, very well given the extraordinarily flexible pricing plans that we've launched with. And again, these will only increase as the tool set increases rapidly. Again, we've also -- are hiring in our Center of Excellence in Bangalore to increase and hit all the road map functionality. And again, I think for the future of drug discovery, co-scientists are really going to be the way that teams work and get the maximum amount of productivity and the maximum amount of knowledge efficiency and use. So again, thank you, everyone, for joining us this morning. We'd be happy to answer questions. Again, you can reach me at ps@lanternpharma, and we'll be happy to do customized demos and give you more details. Thank you very much again. And Reed, thanks a lot for joining.

Yes. Thank you all. There is also a contact form on withZeta.ai, so you can reach us that way as well.

Thank you.

Thanks, everyone.