From Predictive Retrosynthesis to Real-World Routes

Okay. Good morning, everyone. I'm here with Emma Watson from Emma Watson. I'm here with Hermione Granger. Alright. Fourth time. We're gonna get it right. I'll just start. Let's keep going. I wish. Good morning, everybody. I'm joined by Emma Gardner from Millipore Sigma. My name is Sean O'Hare. I'm the senior director of global presales for Dotmatics. Together, we're gonna present, some collaboration that we've done, between our two companies for the past six months or so. So between Luma and the Cynthia retrosynthetic analysis platform. Relatively concise agenda, Emma's gonna start us off talking about, the improving the make of the design may test analyze cycle. I'm gonna pick it up from there and build on that, showing integration and how downstream of that we had some wet lab, dry lab, stories to it. And at the end, if we have time, we'll do some q and a. So the Luma platform from Domainics is I'll talk about this in in some detail in twenty minutes. It's our scientific decision layer. It brings together, all of the enterprise analytical data, application data, and scientific decision making into one, data fabric and interface. Emma is gonna talk about Synthea, so retrosynthetic analysis tool from, from Milibor Sigma. So without further ado, I will pass this over to Emma. Great. Thanks, Sean. So, yeah, I'm gonna be talking about Synthia retrosynthesis software. So first of all, you know, where this kind of fits into your workflow. I'm sure everyone in drug discovery is familiar with the design, make, test, analyze process or the DMTA cycle. Of course, this starts with designing the compounds that you want to make that have the right biological properties or are able essentially affect, you know, your biological target. Once you've kind of designed what compounds that you want, we have to move to the make stage where you actually synthesize them or occasionally can purchase them, but actually get the physical version of the compounds so that you can run those tests, the biological assays to assess, does it have the properties we want, and then analyzing those results to inform, the next cycle. So this is a very iterative process. And the idea here is, you know, the faster that you can go through, the quicker you can get to that hit compound that really has all the right, properties that you're looking for. And the make stage here is one that is often the slowest step. A lot of times, the molecules you're trying to make require multistep synthesis that are gonna be labor and time intensive. So the more that we're able to sort of improve this or accelerate that stage allows you to move through this process more quickly. If we go to the next slide. So if we think about now just within this make stage, there's sort of several steps there, but the first one is really designing the route to make your molecule. And some of the challenges with this pathway design process, One of them just being that, you know, as human chemists, there's a lot of chemistry out there, and, you know, maybe not remembering every possible reaction to do and, you know, kind of needing new ideas, to be able to come up with, you know, really quick ways or elegant ways, to to get to your target there. Of course, you know, kind of looking in the literature to see what's been done before. A lot of times, we're making brand new molecules that have never been made, so there may not be that much literature precedent. Sometimes, you know, try doing a step that's been done before, and it's not reproducible. And, again, needing sort of, you know, another source of ideas beyond just what's in the literature there. And then, of course, you know, a lot of times these are multistep synthesis. Failure can happen at any step, and the later that happens, the more costly it is. So really having, you know, a robust plan at the beginning can help save you, you know, failure later on or or, you know, something that's that's gonna really slow down your process. And then finally, you know, these pathways have to start from something that you can purchase, and this is, you know, again, additional steps. We don't have all the catalogs memorized. So looking to see where you can buy different building blocks. And, again, sometimes maybe missing, not realizing that there might be a compound, a later stage intermediate that you can purchase that could save you several steps, but, you know, just not realizing that's that's available. And so being able to kind of connect all of those things, have a source of new ideas for chemistry, and be able to really optimize your routes can help, you know, save time in that slow stage. Next slide. So Cynthia is able to help with all this to really accelerate that make stage, by being able to do a few things. One, you can use it to essentially screen if you have a lot of compounds, that we're considering in the design phase to be able to filter ones that are going to be really hard to make, and maybe prioritize things that are going to be easier to make, that you can get to more quickly. And then also be able to identify those synthetic routes. What's gonna be the fastest way to get there or sort of the most robust chemistry that, you know, you can do reliably, to get to your target. And then also connecting to the commercially available starting materials so that, you know, those searches are all combined. You don't have to do separate catalog searches and also be able to, you know, identify what things you can get, as well as price information. So all of this allows the chemist to be able to prioritize those easy compounds, help reduce the risk of failure in the lab, and source reliable starting materials so that you can get started as quickly as possible. Next slide. So a little bit about how Cynthia works. So, you know, the key here is Cynthia needs to know all of the chemistry out there, and this is something that's really unique to Cynthia, is essentially how it knows the chemistry. What we have here is a database of what we call expert coded reaction rules. And, basically, what that means is that a team of human chemists has essentially gone through and written, defined the reactions for, you know, all the chemistry that is in the organic chemistry textbooks, you know, that we learn in grad school, all those named reactions and reactions from the literature to define appropriately the scope of what groups can be tolerated by the reaction, which ones are going to be incompatible or require a protecting group, you know, what are the stereo and regiochemistry considerations and be able to really, know, appropriately define that so that when Cynthia is looking at, you know, what reactions can I do here, it's considering all of the nuance that a human chemist would be considering? The other thing they can include in there is information that's useful, like the typical reaction conditions, and some illustrative references so that you have all the information that you need to be able to take this into the lab. And so what I have shown here is just kind of an example of one of these reactions from Synthii. So you can see the reaction scheme is drawn for you, and then the information kind of underneath lets you know this is, you know, a reaction that was predicted from one of our expert coded rules. It's got the name of the reaction. This is a one pot synthesis of the benzodiazepines. And then the typical conditions, And you've got links to those reagents as well as to, some representative reaction references so that if you do want more information there, you can find it. So it's not just sort of, you know, a total black box. You can really see where is this information coming from, and get more information if you need it. Next slide. So, overall, how this works? So as I said, we've got this huge database of all the chemistry. And then Cynthia essentially approaches the retrosynthesis, you know, in a way the same way that we would where you give it your target molecule. Cynthia looks at, you know, what are all the cuts I can make and uses that to generate this complex graph of all of the synthesis possibilities. And this is where, you know, the the computer has sort of an advantage over us of being able to very quickly evaluate, you know, thousands of possibilities that, you know, we might not be thinking of as as a human. And then we have, you know, algorithms that are able to essentially, you know, efficiently traverse this graph to be able to pull out complete pathways and optimize them, for things like cost, step count, reaction type, and things like that. And then what it does is of these, you know, thousands of potential routes pulls out a top fifty that are optimized here, that now the chemist can review, and, you know, find kind of the ideas that are going to work the best, for their synthesis. And so you have down at the bottom here kind of a summary example of a pathway. So, you know, of those fifty pathways, Cynthia gives them to you in a ranked list. You get a summary here of, you know, quickly seeing the number of steps in the path. It gives it a path score so you can, you know, compare different compounds as well as different, you know, pathways for a single target, but sort of finding ones with kind of the lowest path score is gonna be the the the easiest to make. And then you get the summary here where you can see your target, and essentially the the retrosynthetic disconnections and then the different building blocks, and they're color coded there. So you can really quickly see, okay. These are the building blocks that are commercially available. This is how Cynthia's putting my my target together and use that to be able to, you know, prioritize synthesis, plan for starting materials, and, you know, kinda get started quickly. Next slide. And then, you know, you also are able to now kind of sort and filter those results beyond Cynthia's ranking. So, obviously, you know, fifty is still quite a lot of pathways, you know, probably more than you need, but you can use we have, you know, some general filters. You can set limits for price of starting materials, number of reactions, protecting groups, and things like that, to kind of adjust the list, as well as more specific things. So if there's a specific, you know, starting material that you don't wanna use or an intermediate you're trying to avoid going through that, you know, maybe, you know, is probably gonna be difficult to purify or something like that, you can exclude specific things and prioritize the chemistry that you like. And then you can also see, in this screenshot here, you get, you know, a list of of really the the path the full pathway details, of, you know, the full reaction scheme there. Next slide. And then, you know, then that next step is once you've kinda picked out a pathway that you want, you've got quick links to all of the commercially available starting materials. We do connect to around fourteen million commercially available compounds, unique structures there. So, really, you know, covering a a wide synthesis space. We have around ninety different suppliers that we work with, And these are all, you know, reliable suppliers that we've vetted of, you know, real compounds. So it's not something that's going to be, you know, a virtual compound that you may or may not be able to get. These have all been vetted to be, you reliable suppliers there. And what you can also do is connect and see kind of directly, the pricing and availability. You get, sort of a summary price. It's that pink number on that you can use to compare, you know, what things are cheap and what things are are more expensive. But then, you know, you can kinda connect to see the the real pricing and availability. So, like, in this example here, you could see, okay, I can buy five grams of this for fifty eight dollars, and it's, you know, available to ship right now. And then you can use that to to make a a shopping list that you can either, you know, purchase directly or, export to kind of keep track of these are all the the things that I need. Next step. Awesome. Okay. So, you know, the main Cynthia interface is a web app that you can access directly and, you know, do all of that kind of standalone. But we do also have it available as an API or application programming interface, to be able to connect with your other chem informatics tools. You know, we know that the the synthesis planning is just kind of one piece of this process, and there's a lot of other data that you wanna be able to to look at together to really, you know, inform your planning and and look at sort of the the full cycle there. And so, this is something that we've been working with with Luma, to kind of directly connect Cynthia in with your other data, to be able to give you a more, you know, complete picture in a single place. And so to talk more about, Luma and that connection, I'll I'll pass it back over to Sean. Thank you, Luma. Let me talk about Luma in general for a couple of minutes, and then I'll show the integration, and then we'll do, an example of wet lab plus dry lab at the end. So legacy laboratory system, we've all sort of dealing with this now in some form or another. You've got things everywhere. A lot of Excel, a lot of different systems, not as much paper as we used to have, but not too long ago, but but information is sort of everywhere. So there's a legacy of silos. You gotta go ahead and collect data, pull it together, do, data wrangling, all these other things, make more Excel sheets from other Excel sheets, do everything in the same place. The mission of Luma was to create a a fair data fabric, and inter application interfaces, around those data. So we took a sort of an unstructured world and brought those things together. A lot of enterprise value from this, you know, silo data, better collaboration, less data loss, all these other things, less admin, less data wrinkling, and those types of things. So what Luma kind of looks like at a high level is a number of, ingresses. These are, the icons for a bunch of our tools. This is RELN. There's a piece called Luma Lab Connect, which allows us to bring in instrument data. And then this plus sign down here, which is any other thing that you can integrate with, with, other systems through API or database connections, and this is where Cynthia fits in. So using Cynthia's API set, we integrated Luma to basically recreate a Cynthia interface and and use that for downstream, activities. So ingress patterns for for Luma files, API, database connections, the API is how we how we did this with Cynthia. This comes into play a little bit later again, so I'll just bring it up now. We started this off with this design, make, test, analyze cycle, and we're gonna be talking mostly about the make here. And I'll I'll mention test and analyze, and we'll we'll leave design for maybe another another session. So what Cynthia and Luma do together is they combine the power of in silico retrosynthetic analysis, which I might just spoke about with in house experimental expertise. So this is your ELN, your LEMS, whatever the tool you're you're on, your lab automation systems, whatever those things are, and then then allow us to do dry lab wet lab comparisons later on. So I'm gonna show the the the demo. So what you're gonna see here is essentially a fairly fast Cynthia demo minus a lot of the details that Emma just showed you. And it's it's a video. It takes a couple of minutes. So here, if I put a molecule in, this is rosuvastatin, which, some of us may maybe take, submit that to to Cynthia. This is I'm gonna pause this quickly. This is an actual, interface that lets us kind of tune this. I've I've made it deliberately broad. So I've said you can spend up to ten million dollars and do two thousand iterations. So it's gonna go and do a bunch of work in Cynthia, pass that through the API set, and do the actual, representative analysis. So it sends this through, creates a request. These are data that are actually coming back from Cynthia in the Luma app, Luma interface. So there's the the submission I just did. It pulls back statuses from Cynthia itself so you can kinda, you know, maybe pop up and get a cup of coffee here, come back, and it eventually reads success. So there's rosuvastatin. That's the results that we just made. Click on that, and I get the list of all the pathways. Let me pause this again. So where the the cursor is on the screen now is that's a list of the pathways there. There's probably r fifty. As as I said, it kind of cuts off there. A number of data along the top, so I get a pathway score. This is sort of the the feasibility or the the the goodness of the route, number of reactions, and they wanna filter this out to less than eight, whatever it is, and some other information as well. So all that stuff gets passed across as well. I can click on a pathway or pathways and see that in a, sort of a graph view. Sorry. Here's the graph view itself. I can zoom in on this. And then as Emma mentioned a minute ago, I'm a synthetic chemist, she does as well. We do know a lot of chemistry, but we don't know all of them. So Cindy actually gives you the the name reaction as well, and you might wanna click in there and get some more information. So using, an AI tool, I can click on the Hansen's Apparels, gives me a description of that chemistry. So I make run across a reaction for her to before and say, what what is that reaction? Go ahead and get some information on that reaction mechanisms and references, significance, etcetera. I can click on any of the notes, get information about the molecule that was created, and so on and so forth. So there's a lot of information that can pull, out of the results of itself and then, use AI on top of that to get more information. Alright. The next thing I can do is I can look at a particular pathway and pull up the the view that that Cynthia creates. You saw this on Emma's slot. I did not do that. Let's rerecord that whole section, Ryan. Sorry. Let me back up. I'm just gonna start from from this click. So I'll do this, the whole demo itself again. Sorry about that. And I'll take my hand off the mouse because apparently, it's very sensitive. Okay. So what you see here is the Cynthia interface in Luma. So this is rosuvastatin. I click submit. I get a dialogue here that actually lets me tune the API call. I've kinda made it some big assumptions to to get me a bigger set of of results. That creates a request. That's gonna go off to Cynthia itself, and I'm gonna get some dialogues back and say, okay. It's working on this. So in a second, I'll see the request show up. And then down below, I'll kind of see the status of that request over over a couple of minutes, I'll fast forward a little bit as we go through. So there's rosuvastatin, the request that is made. I now see it down below. It's in progress. These are actual updates from the Cynthia application itself. A couple minutes later, it's done. So I've gotten kinda, got a cup of coffee. I've come back. I now have my results set, so I'm going to look at those. So here, I see the list of of pathways that's on the left. The spot shadow here is showing you the information you get back. So I get pathway scores. That's, something that I mentioned, reaction counts, molecule counts, things I can filter out later if I want to kinda choose different, a different subset. Down below, I get the graph view of the pathways I just took. I can zoom in on these things. Emma mentioned that, you know, even as synthetic chemist, we don't know all these reactions by name. We don't know what they all are. You might come across a weird one. And since this is not particularly weird, but I can go ahead and click on that and get description from the AI says, okay. This is what this actually is. Here's some key some key details, reaction mechanism, all of these things that I can I can tune this as well, this is giving me more information on the routes itself? I can click on any molecule within the route and get information on it. And so I can I can interact with this a little bit and get information, to kinda help me make my decisions as I go through? The next thing I can do is I can click in on the route itself, in that pathway selector. So let's say I wanna see this one. I wanna see the the actual Cynthia view of this actual synthesis with the reaction steps, links to, catalog purchasing. So it's it's a bit bit zoomed out. I can zoom in a little bit there. You can see those little flasks are actually links to the catalog if I wanna buy something. I can use again the the AI to describe this this entire pathway. So it says, okay. Here's what you're doing. Here's the steps. Here's the reagents involved and so on and so forth. So I can get a bunch of more information on top of this. In some of these, you might get actually a safety warning, is kinda nice. This reaction is exothermic, whatever it might be. And that's, in a nutshell, the Cynthia integration, on Luma. So what we've done now is run the the Cynthia query, pull the information back into Luma, and now we can do the value adding part. So as a chemist in the lab, I go ahead and I do I would do that exercise either in Cynthia or manually. I'd go ahead and I'd go to the literature and I pull this information back, and I'd make some decisions on what I wanted to do. And, historically, you know, we've had a number of ways we can do that. We can go search in ELN. We can do more literature searching, a bunch more human work to kinda refine that list. Now with Luma, we can go ahead and say, let's take the retrosynthetic analysis that we just created, and let's compare that to what we actually know how to do in the laboratories, what we have equipment to do in our laboratories. And to do that, as a person, would take me quite a bit of time. I need to go and I need to find some people. I need to some ELN searches. I need to pull this all out and and combine these things in my head. What I'm gonna do instead is I'm just gonna ask the the Luma AI agent a question. I don't know the data model exactly. I don't really know what all these things look like, so doing a comparative search would be maybe challenging. The Luma AI agent was actually released this week, so it's kind of, kinda new to the market. I'm gonna talk about it a little bit here. And all I'm gonna do is is type in a prompt. Okay. And I'll I'll pause it when the prompt comes up. Alright. So, basically, I don't know what data I have. I don't really know who's who. I don't know anything. So I've I've just done a a registered synthesis of rosuvastatin. I wanna go and compare that with everything in my chemistry ELN. I'm just gonna ask it. Go ahead. Look at the chemistry ELN. Give me some insights on which pathway I should choose, which I should avoid, who my local experts are. So it's gonna find a whole bunch of useful information for me. And then what do I need to buy, and and can I buy it from from the Leopore Sigma catalog? Alright. So I'll let that run. Again, it's just a just a prompt and any configuration for this. I've just asked a question. So it goes through. It starts this off and starts starts running an agent agentic AI starts pulling back information. Alright. I'm gonna cut to the chase. I'll take along the chase towards in a second, and I get a report. What are the recommended pathways it says I should do? Which one should I avoid, and why? In this case, I think it was mostly on cost. In some cases, it'll say you don't actually have the equipment to do that or you don't have any expertise to do that. Here's our local experts that go find these guys to ask them questions about this. These are the commercially available materials from the catalog. Critical, reagents from ELN experience, critical lessons from ELN, all of these things that are combining wet lab dry lab. They're really simple. Again, just a question to the data, using the agent. So that took six minutes and twenty one seconds to do that. I think it would have taken me a day, at least, to do it actually manually. So I wanted to kinda showcase this is what we're we're bringing to, scientists, to our partners as we collaborate and and are able to add, combined workflows between different tools. Alright. So that's make wet lab dry lab. As I mentioned earlier, we can also bring in test results. We have the the same ingress patterns I showed you before, files, APIs, instrument data, database connections, etcetera. Okay? And then we can go to analyze. So all these data are on the same data backbone. So we can generate reports of data. This isn't the same data I just showed you, but, but by combining, test data, synthesis data, various analytical results, I can start to build out dashboards and show those those things. I just want to finish the story. This we're not gonna go too much detail here. Again, this might be a story for another day. So to kinda summarize what we just talked about, the the Synthia retrosynthetic analysis software allows scientists access to a wide breadth of synthetic information. They can put in a a molecule, get enormous amount of information back on how to make that molecule. By combining that with dotmatics Luma, we can leverage all of those capabilities, combine those with any other capability you have in your organization, whether it's your ELN data, analytical data, downstream information, and start to build out richer applications and solutions, with our with our partners. So to summarize here, a unified work workflow between wet lab and dry lab, context aware routes. So we've got an enormous amount of retrosynthetic capability enriched with context from actual experimental data in drone laboratories, AI assisted decision making, and hopefully, that leads to, fewer cycles, better decision making, lower cost, faster time to market, all of these things. With that, I'll conclude. And if there are questions, we're here to answer them. Thanks, everybody.