LiveDesign for Biologics: Uniting Data, Tools, and Teams

0:06 

Thank you so much for joining Schrodinger's workshop today on computational approaches for engineering peptides and proteins with enhanced properties. 

 
0:14 
I am Cindy Gerson. 

 
0:16 
I am a Lead Product Manager at Schrodinger working on the Enterprise Informatics platform, specifically working on adapting the platform for supporting biologics, which is what we're all here for. 

 
0:28 
So in this workshop, we're going to be discussing these approaches from both an informatics perspective and the computational modelling, physics based modelling approach. 

 
0:39 
So I hope that this will give you a holistic view about how you can use tools on both sides to  help you expedite and improve your workflows. 

 
0:54 
So the agenda today, we're going to break this up into three sections just because it is a long 2 1/2 hour workshop. 

 
1:01 
We want to make sure everybody says stays, you know, rejuvenated, awake and aware. 

 
1:07 
So in the first section, we're really going to talk to you about our enterprise informatics platform. 

 
1:10 
My colleague Noj, who's standing next to me will introduce himself, he is going to give you an introduction to the live design platform as a whole, it's overall approach. 

 
1:19 
I will then jump in to give you an overview about how we've been working to adapt this platform to support biologics. 

 
1:25 
And then we're going to jump into an example workflow, how we're actually using this platform to perform a common workflow like biologic candidate triaging. 

 
1:34 
In this case, we're going to do a peptide triaging workflow. 

 
1:37 
For the first part of this workflow, we're going to talk more about how you would analyse and triage biologic candidates using experimental data and embedded platform functionality. 

 
1:47 
Then we're going to take a break and when we come back, we're going to talk more about predictive approaches, physics-based modelling, approaches that can give you predictive parameters. 

 
1:56 
And my colleagues Dan and Yelena will give you some background on different approaches that we have at Schrodinger for doing so. 

 
2:04 
And then I will jump back in and show you how you can take all of the wonderful tools that they're showing you, all of the wonderful predicted properties that can help guide your workflows and actually integrate those into our informatics platform so that you can have wet lab and in silico data side by side as you're developing your work flows to make them as efficient as possible. 

 
2:23 
We'll take another little break just to, recheck the minds. 

 
2:26 
And then when we come back, we want to give you a flavour that this is not just for peptides. 

 
2:31 
Our example workflow that we're doing in depth to give you a real sense of these platforms is a peptide workflow. 

 
2:36 
But we're going to show you a couple of antibody-based workflows as well with both, you know, platforms being integrated together. 

 
2:42 
One again, a triaging workflow and then the last one talking about how you could envision doing a mutagenesis optimization workflow platform. 

 
2:52 
I'm going to pass it along to Noj now to give you an introduction of the informatics platform overall. 

 
3:00 
Thanks, Cindy. 

 
3:01 
Yeah, so my name is Noj Malcolm. 

 
3:03 
I've been working on the LiveDesign platform primarily in small molecule world for the last 10 years. 

 
3:10 
This is something that we've been developing in collaboration with some pharma partners, but very broadly used. 

 
3:20 
So pretty much half the top ten, top 20 pharma are now using it, a large numbers of biotechs. 

 
3:26 
And the philosophy that we tried to drive here at Schrodinger historically has come from predictive methods, physics based modelling world and starting to think about how we can enable the people that are actually doing the design, doing the science to get easier access into those tools. 

 
3:47 
And that sort of was the genesis of the LiveDesign platform. 

 
3:52 
Cindy will talk about the triage components. 

 
3:55 

We don't do these things in a vacuum. 

 
3:57 
We're doing these in the context of the existing experimental data. 

 
4:01 
So this is a really important part of how we have a platform and what we can do with that platform in terms of getting everyone on the design teams, everyone that's working on a project, access to the same sorts of data, whether that's the in vitro data or in silico data. 

 
4:23 
We built this platform very much in a technology and science agnostic way. 

 
4:30 
So the idea being that we can plug in components that are bespoke to different customers, different workflows they're doing. 

 
4:38 
We all have very similar things, but similar but different. 

 
4:42 
It's the way I often think about it. 

 
4:44 
So we're looking at the design tools that people are using or computational tools that people use, and those may be slightly different. 

 
4:51 
And we think of this often in this sort of context. 

 
4:56 
We see sort of three key components, I think, to what we need to do to be able to enable researchers across the labs. 

 
5:06 
So bottom left piece of digital continuity. 

 
5:09 
So this is what I was talking about, getting access to data. 

 
5:11 
This is a non-trivial problem, getting people to actually be able to ask the questions of the data that they already have so that they can think about where they are in the current stage for project design. 

 
5:24 
And being able to also in that context, capture the knowledge, capture hypothesis. 

 
5:29 
Why are we trying to do this? 

 
5:31 
We're trying to work in a scientifically intelligent way when we're doing design, hopefully nothing. 

 
5:36 
We're not just doing things by serendipity that we're actually able to capture hypotheses and test those hypotheses as we cycle through projects and then learn from that and come back and sort of drive faster hopefully cycles as we move forward the bottom right hand side. 

 
5:56 
We see machine learning AI exceedingly fashionable at the moment, Schrodinger definitely is exploiting these technologies, but we have a very strong belief that the physics based methods can really help drive not just projects onto themselves and offer insights there, but can help drive those ML efforts. 

 
6:16 
So being able to capture and exploit those in the same context as our experimental data is something that we see as exceedingly important. 

 
6:27 
And then sort of top bullet here, which is not something new, but I think it's growing and growing the more we see this in the market, externalisation and collaboration. 

 
6:37 
So the way that customers companies are working now is optimised. 

 
6:45 
You know, they'll be focusing on the components that they're greater and they'll be working with partners outside. 

 
6:51 
So whether those are collaborator partners or they're CROs where they're just doing work, but having a platform where you track that work, the things that have been sent out to CRO or work, collaborative partners, LiveDesign is something that's been used extensively in that context as well. 

 
7:16 
We're really looking at LiveDesign. 

 
7:22 
Yeah, from my perspective, I've been working for the last 10 years on small molecules. 

 
7:25 
So we've been building up this platform. 

 
7:28 
So I think this is one of the exciting things about what we're doing. 

 
7:30 
It's not a new platform that we're building in the biologic support for. 

 
7:35 
It's the same one. 

 
7:36 
A lot of the functionality and a lot of the tools that we need as we cycle around this classic DMTA cycle are the same. 

 
7:44 
I need to get access to data. 

 
7:46 
Well, that doesn't matter whether it's a biologic or a small molecule or an OLED or a formulation, you know, the actual data access is the same problem. 

 
7:55 
Analysis of the data, being able to build out dashboards, drill downs in terms of that data plots, etcetera. 

 
8:03 
Those are all the same things. 

 
8:05 
Being able to execute computational predictions so that we can actually take a predict first philosophy and things, again, it's the same. 

 
8:16 
There's differences, different entity types, the computations we're running a different, but as I said before, we have this agnostic view. 

 
8:23 
So actually it didn't matter really when we did that. 

 
8:27 
And tracking what we're doing. 

 
8:30 
So you know, as we're making things potentially internally, potentially with external collaborators, those are the same tools really that we need to do all of those things. 

 
8:42 
LiveDesign is a web-based platform. 

 
8:45 
So we'll be seeing some live demos. 

 
8:47 
Hopefully the Wi-Fi down here still holds up. 

 
8:52 
And that's a really important part for us, particularly as we're looking at deployments throughout organisations. 

 
9:00 
IT cost of ownership is an important consideration when we're looking at what we're doing here. 

 
9:09 
So this is an example for the small molecule side of things. 

 
9:13 
So as I mentioned Schrodinger has historically been very strong in the physics based modelling. 

 
9:21 
So we're looking after the structure based information, but being able to take laboratory information alongside predictive tools and compare those so that we can drive cycles is a fairly powerful way to enable design programmes. 

 
9:44 
The agnostic way that I mentioned earlier is shown a little bit here in that we have collaborations with plug in other tools, whether those are predictive tools, whether those are in house. A lot of the people we're working with are all already doing their own AI/ML workflows. 

 
10:05 
So being able to include those and expose those to bench scientists is very important. 

 
10:11 
Working with third parties. 

 
10:12 
So for example, with TIBICO Spotfire, they're doing more advanced analysis or plugging in those tools. 

 
10:18 
And what we'll talk today about some of these things, some sort of internal words that we use. 

 
10:26 
LiveReport is fundamentally a spreadsheet of data. 

 
10:30 
It's a view on into a set of data. 

 
10:33 
And that may be a set of entities, it may be a live query. 

 
10:38 
So we have capabilities to create views of data that update as the data environment that you have changes within your project as it's evolving. Models conceptualise effectively columns, anything that we can run command line, web service, whatever we can integrate there. 

 
11:00 
We have a large number of Schrodinger tools that we have included into the workflows that we'll be seeing later on today. 

 
11:10 
But it could be absolutely anything. 

 
11:13 
We have this concept of these forms. 

 
11:15 
So it's not just a spreadsheet. 

 
11:17 
And again, we'll see that in some of the examples today about how those allow us to build up rich analyses to build custom views into the data for the specific project of the team within the project about what they are looking at. 

 
11:33 
So it's not a fixed workflow. 

 
11:35 
We can customise those workflows very easily and I said we can plug in any of these sort of external technologies, internal, third party, etcetera. 

 
11:47 
The platform is fully exposed through an API. 

 
11:52 
So that allows it not just to be a place where the design is done, but it allows it to interact with the rest of the infrastructure within an organisation so that that information that's captured within LiveDesign can be used elsewhere. 

 
12:06 
So plug it into the ELN or lab notebooks, whatever that might be. 

 
12:14 
And then just at the end. 

 
12:21 
Just to touch on a little bit about where we see tools here. 

 
12:26 
So the design, we believe very much philosophically in this predict first view on data. 

 
12:33 
But when we're making things actually capturing what is the stage of the small molecule or maybe a synthesis. 

 
12:41 
But where are we, what is the tracking the times on that, trying to look at telemetry on projects etcetera so that we can help to optimise those. 

 
12:50 
And then look in once we've got experimental data, how can we  look at that? 

 
12:55 
We want to see our DRC curves etcetera and then more advanced analyses using various graphical tools. 

 
13:06 
So that's sort of a history of where we've been coming the last 10 years. 

 
13:11 
And hopefully I think we've been pretty successful in the small molecule world. The last three years we've been on a journey to try and move towards a biologics framework. 

 
13:23 
And the first stage that we took on that was trying to abstract away a dependency on working on small molecules to have what we would call a generic entity, so that LiveDesign could work on any type of entity. 

 
13:34 
And that's where we are today. 

 
13:36 
And the work that Cindy's been driving for the last couple of years that we'll talk about is how we actually make LiveDesign understand the nature of the actual biological entities we're working on. 

 
13:49 
And so with that, I'll hand over to Cindy. 

 
13:54 
Thank you, Noj. 

 
13:56 
So that gives you an overview of LiveDesign as it exists, the generic view of the platform. How it could work for small molecules, how it has worked for small molecules for so many years. 

 
14:09 
But what I've been focusing on since I've joined Schrodinger a couple years ago is really making sure that we enable biologic support in the platform so that we're developing the platform to take advantage of everything that's been built foundationally, but make the platform biologics cognizant so that it understands what biologics are. 

 
14:26 
It gives you the tools that you need to deal with them and to deal with their workflows and make all of your processes more efficient and effective. 

 
14:35 
So a lot of the core functionalities that Noj just described to you are really critical for biologic support as well. 

 
14:43 
So there's a lot of those foundational principles that are going to be extremely impactful right off the bat. 

 
14:48 
So prior to my time at Schrodinger, I worked in biologics discovery for 12 years and I saw the informatics challenges first hand. 

 
14:57 
Very often you're dealing in a world where you have all of the pieces of your workflows, all of that data, all of that metadata, all of those process decisions trapped in independent silos and repositories where all of your data is stuck in the step that it is in. 

 
15:11 
And often it is difficult to have commutative information between data that you have from early stages in your process with output data from late stage and process. 

 
15:21 
So if you're trying to actually connect information from each step in the silo, it's very difficult. 

 
15:28 
Often your tools that you use, whether it's analytical tools or predictive modelling tools, those are often disconnected from your data silos as well. 

 
15:36 
So every time you're trying to use one of those tools to help you improve things in your workflow, you're very often having to jump in and out of the platform. 

 
15:46 
The idea of LiveDesign and what it has been doing for the small molecule world for so long is to get all of that under one hood. 

 
15:53 
All of your data silos from each stage of your workflow process, all of your analytical tools, all of your predictive tools, all under one hood. 

 
16:00 
So everything is hooked in, interconnected and you can have seamless workflows. 

 
16:05 
And this a call back to what Noj showed you for small molecules. 

 
16:09 
Again, the similar idea that if you have biologics in platform, you should be able to have computational assays, wet lab assays from all stages of the process side by side. 

 
16:18 
So you have all the information you need to make the best decisions possible. 

 
16:22 
So a lot of what live design has already been for many years is already getting us let's say 80% of the way to what we need to be able to support biologics discovery and development in platform. 

 
16:33 
That being said, we completely understand biologics are different than small molecules and they are going to have different needs for both their workflows and their analysis needs. 

 
16:43 
So we will need to make platform adaptations and we've been working very vigilantly for the last couple of years to do so and we'll continue to keep building on that progress. 

 
16:52 
If you look at, a small molecule DMTA cycle like the one that Noj already showed, it's really this, you know, very predictable loop of design, make, test and analyse with a very design first approach. 

 
17:03 
Design, use predictive tools, analyse those predictions before you even get to the point of making and testing. 

 
17:11 
So that allows you to have fewer cycles and much faster discovery processes. 

 
17:16 
The biologics discovery cycle, and this is just an example, everybody's biologics discovery cycle looks different, even if it's for the same modality. 

 
17:22 
And obviously different modalities have very different cycles. 

 
17:25 
This is kind of an example of what an antibody design cycle could look like. 

 
17:29 
It is not this simple cyclical loop. 

 
17:31 
Very often, you're starting off just not in a design first, but very often in a make discover first mode, creating these large libraries upfront and doing a lot of production and testing before you even get to the point where you even would think of predictive tools for optimising your biologics. 

 
17:49 
This creates a lot of data and metadata from every stage of this process to capture. 

 
17:54 
And it's really important to get all of that information under one hood, both for being able to analyse your biologic candidates efficiently, but also for being able to capture those decision making processes, the why and the how you move things from one stage to the next, because that'll allow you to also be able to do process refinement as you start to capture those decisions and what choices actually led to best in class therapeutics. 

 
18:18 
But the really interesting part is once you get to the testing phase, you don't just go in the cyclical loop or you loop right back to the beginning. 

 
18:24 
Very often what you see is guiding where you move back in the process to do optimisation and the ability to track those divergent and disparate workflows where you might be moving in six different directions based on what you see is something that we need to be able to accommodate in platform. 

 
18:41 
Additionally, biologics, unlike small molecules, biologics is a huge umbrella. 

 
18:45 
It's not just one thing, it encompasses many different most modalities, nucleic acid based, amino acid based and some, some of the more smaller scale almost bordering small molecule, larger scales like antibodies, TCR. 

 
19:00 
And we are trying to make sure that the tools we put in place will allow you to handle the discovery and development cycles of all of these modalities and all of the workflows and the complexities that come with them. 

 
19:11 
And we know that there's some biologic specific analytical tools. 

 
19:14 
Just a quick example, there's no sequence that's associated with a small molecule. 

 
19:18 
We need to be able to have sequence analytic tools and other biologic specific tools hooked in and accessible in platform so that you can do your workflows without needing to jump out somewhere else. 

 
19:28 
Because again, the whole principle of LiveDesign is everything under one hood, one stop shops, centralised workflow hub. 

 
19:37 
So as we've been building LiveDesign for biologics, we really think of these guiding themes. 

 
19:41 
The first bit is really about being able to enable workflows that are critical for you. 

 
19:47 
The three big buckets of workflows that we're looking at are being able to allow you to have in platform expedited and improved biologic candidate analysis and lead selection. 

 
19:57 
Again, this is really around getting all of that data, whether it's experimental or in silico in one platform with all of your analysis tools hooked in so that you can do everything in one place and that'll make everything way more efficient. 

 
20:10 
The next you know workflow buck is around have being able to optimise those loop back steps. 

 
20:16 
Being able to loop back in platform, say I've gotten to this stage, I want to loop back to this new workflow to optimise and ideate new things in platform, get predictive information upfront about them and decide whether or not I even want to make them. 

 
20:28 
And the last workflow bucket we're really thinking around is being able to have process improvement. 

 
20:32 
Being able to capture all of those decisions, the whys and the hows of why you didn't. 

 
20:37 
How you went through this process so that when you have that information hooked into the platform, all of those pieces of information on decisions, you can start figuring out what decisions actually led to best in class therapeutics so that you can start making your processes more efficient and wasting less time on methods that actually weren't as productive. 

 
20:55 
So as we're working through being able to enable these workflows, you know we're going to build this on piece by piece. 

 
21:01 
Rome wasn't built in a day, but these are the things we have in the back of our mind in terms of workflows. 

 
21:05 
We want to make sure we enable to support all of you. 

 
21:08 
Again, there's not just one modality in biologics. 

 
21:11 
We want to be able to support amino acid based and nucleic acid based things containing natural and non-canonical monomers of a whole wide variety. 

 
21:19 
And we're working to build out the number of modalities that we really can handle in platform and handle their needs and supportive tools and functionality. 

 
21:28 
I made a brief example in the last slide. 

 
21:30 
Obviously you have to be able to look at sequence if we're looking at biologics now. 

 
21:34 
So there's a lot of biologic specific tools that we’re looking to build out that were not needed for small molecule are very critical for biologics workflows. But also general data and workflow functionality that even though it hasn't been built in platform yet and it will be incredibly useful for biologics. 

 
21:54 
It's not specific for biologics and will be helpful to the small molecule in the materials world as well. 

 
21:59 
So there's of plans to build out all of these functionalities. 

 
22:05 
And we're, working towards that for right now, the current status of biologic support and LiveDesign, because we can't do everything all at once, is really focusing on this first workflow and being able to to get all of your data in one place with some biologic specific tools that will allow you to go through workflows around improved biologic candidate analysis and lead selection. 

 
22:29 
So the initial offering, and we're going to be showing you this in a live demo today, is all around focusing and enabling those kinds of workflows. 

 
22:37 
And really the information you need to be able to get into this platform to perform these work flows. 

 
22:42 
Are pieces around what defines the biologic, what the biologic is, it's sequence, it's structure, it's format, as well as getting all of the information in that defines this biologic’s potential as a therapeutic. 

 
22:54 
Having your experimental data from every stage, not trapped in silos, but now in every stage of your process, end to end in the platform. 

 
23:01 
Having your in silico data, any predictive models inside the platform so that you have a view onto all of the information you need in one spot. 

 
23:11 
This is a list of the specific functionality, how we're addressing that need to get what defines the biologic and what defines its potential in one platform. 

 
23:20 
We have enabled biologic specific registration. 

 
23:23 
It's compatible with biologics that are defined by both monomer strings or HELM. 

 
23:28 
I know peptide folks in the room, a lot of them are very big on HELM. 

 
23:31 
We have put that into place and to give a visual representation of what the biologic is. 

 
23:36 
Again, its a cognizant platform, when you give the biologics to the system, it understands this is an antibody, this is a peptide, this is DNA, it has these chains or strands attached to it and that's the sequence that's associated with it. 

 
23:51 
Again, just touching on HELM functionality, giving people who are ingesting molecules that way, the ability to have beads on a string representation and have custom monomers. 

 
24:00 
We don't all work in natural monomers, non-canonicals are ever growing. 

 
24:06 
Having sequence view and analysis tools in platform, being able to search and filter for biologics based on sequence motifs. 

 
24:14 
Obviously that's not something that we had to address with small molecules. 

 
24:18 
But if you want to be able to say I want biologics with this sequence motif in this chain or in this region, being able to pull out a list like that is something that we wanted to enable to make your search and filter functionalities work best for you. 

 
24:31 
This is a feature we refer to as composite row. 

 
24:34 
I'm sure we will show bits of that at some point. 

 
24:37 
But the idea is if you have a biologic, especially a large biologic that's made of multiple chains, multiple components, being able to access those individual pieces. 

 
24:46 
This is really critical for ideation. 

 
24:48 
So if you have let's say a multi-specific, being able to breakdown, see those individual pieces, have those pieces identified. 

 
24:56 
So if you want to grab them as you're evolving a new molecule later on, you have access to those pieces as well as the whole biologic. Getting all of that data in, data and metadata associated with your biologic in platform. 

 
25:10 
Being able to execute models in platform and be able to visualise these, especially if you have 3D predictive models, being able to visualise them in platform side by side the way that Noj showed you for a small molecule with that beautiful rotating structure. 

 
25:23 
We want that for biologics as well. 

 
25:26 
So those are the pieces that we really worked over the last couple years to put in place to enable this first bucket of workflows. 

 
25:34 
And with that, I want to show you an example because I can sit here and talk, you know, wax poetic on all the functionality all day, but it's really seeing it that gets you to see the power and it's the potential. 

 
25:45 
So the workflow that we're going to be showing you today, it's an example of how we put all these pieces together to help make your processes more quick and efficient. 

 
25:55 
Is a triaging workflow of p53 mimetic peptides and basically p53 is a tumour suppressor protein. Ithas lots of negative regulators that want to stop it from doing its job. 

 
26:07 
MDM2 and MDMX are two key ones. 

 
26:11 
And the idea is we want to develop these p53 mimetic peptides that can bind to MDM2 and MDMX, preventing them from binding to p53 so that they can't negatively regulate p53's function. 

 
26:24 
There's been lots of work on this over the years to develop the best p53 mimetic peptide that really allows it to act as therapeutically well as possible. 

 
26:33 
And we're going to go through a triage and workflow is if you had a whole huge set of all these mimetic peptides, how could you actually use this platform to make that triaging process as easy as possible? 

 
26:45 
So we're going to start again. 

 
26:46 
We're breaking down this workflow into a few steps. 

 
26:48 
I'm going to start with just getting you into the platform, showing you what it is and get looking at it from how you would deal with the experimental data in this platform. 

 
26:56 
Slicing and dicing the data, applying different, visual tools to help you see more easily, be able to discern a more promising candidate from a less promising candidate and use certain embedded tools. As Noj said, we have a lot of tools that are directly translatable to be useful in biologics workflows that were already there for small molecules, things like a multi parameter optimisation. 

 
27:19 
So being able to take many parameters that are important to you about what qualifies this candidate, distil it down to this one number to rule them all. Having that as an option on the platform. 

 
27:30 
Gives an easy, quick visual of how this candidate performed using what we call free form columns, ways for you to be able to communicate in platform and Forms views, which Noj referred to, which are really just custom-built analysis views. 

 
27:46 
And it's the Forms views I show you today. 

 
27:48 
These are not fixed in stone that if you go into this platform, you don’t have to use that analysis view. 

 
27:55 
These are all able to be built out individually so that you can look at the data and what visualisations you need to analyse the data, however you want to set it up. 

 
28:05 
That's the real thing. 

 
28:06 
I want to drive home when I show you the forms you use. 

 
28:08 
I built mine. 

 
28:09 
If I if Dan or Noj or Elena wanted to build a different one because they want to look onto the data differently perfectly doable. 

 
28:15 
This is not, you know, set in stone as I'm showing this to you. 

 
28:19 
So with that, I'm going to jump out of here and I think I have to do this so I can switch over. 

 
28:26 
I'm going to jump into Livedesign so that you don't need to just hear me talk, but you can actually see what we're discussing. 

 
28:32 
I'm going to just expand this a little bit so that pictures get a little bit bigger. 

 
28:40 
So where I'm starting off right now, I tried to start off in the most empty place that I could because I want to show you how you can actually build these views, whether in the more spreadsheet mode of a live report. 

 
28:53 
And this is a live report, by the way, this is your view onto the data. 

 
28:56 
Now, this is not, you know, a spreadsheet. 

 
28:58 
This is a view onto the data. 

 
28:59 
So if you're hooked into your various databases where you store information about your biologic, what it is and what defines it, experimental data, whatever it might be, we are hooking into those and we are giving you a view onto them here. 

 
29:11 
This first mode is just kind of familiar looking and that it does look like a spreadsheet, but don't think of it as a spreadsheet and that it's like a static. 

 
29:17 
I cut and paste my data into your It's a view onto your data. 

 
29:22 
So again, starting the most basic place, I want to show you how you could build this up and how you can add in all of the different types of data that you would want. Slice and dice it in different ways and create different analytical views to be able to triage your candidates how best makes sense to you. 

 
29:37 
So here is a group of peptides. 

 
29:40 
We have a 214P53 mimetic peptides in here. 

 
29:44 
You can see as I scroll through and what you'll notice is you have this beads on a string representation of them. 

 
29:52 
And again, this is about when you have a smaller biologic and you just have a series of you know, 10-20 monomers. 

 
29:59 
This beads-on-a-string view of monomer shows not just the monomer identities, but how they're interconnected, especially if you work on things like cyclical or branched peptides, being able to view that interconnectivity and see how it actually maps out is really nice to have in platform as well. 

 
30:15 
So these peptides, you see that pretty picture of what the monomeristic view looks like. 

 
30:21 
Does it actually understand that that picture is associated with your biologic entity? 

 
30:27 
It actually is the sequence and being able to understand and extract that sequence to analyse it. 

 
30:32 
So I'm going to pull up for you just really quick our tool that we put our sequence viewer tool that we put in platform. 

 
30:38 
So you can see that it knows that this is a peptide and that this is the sequence associated with it. 

 
30:46 
And if you can see, we have a tool tip. 

 
30:48 
If you hover over the monomers in it, you will get some information not as exciting when you're working with your standard flavours. 

 
30:57 
But if you have non canonicals as part of it, you can hover over and get a lot of information what the actual monomer is, what is its natural analogue, any properties you're interested in. 

 
31:07 
So we have this and what you'll see is as I click on these, you're automatically updating in here the view on sequence. 

 
31:16 
So it knows, it is biological cognizant, it knows this is a peptide, it knows this is the sequence associated with it and it can display what that is. 

 
31:25 
We have some, you know, fun tools kind of built in here. 

 
31:28 
The ability to do either multiple sequence or pairwise alignment. 

 
31:32 
The multiple sequence alignment isn't super impressive for peptides that are all the same lane. 

 
31:35 
So it but if you see some of the antibody work flows later, you'll see how that alignment and platform is really useful. 

 
31:43 
We have the ability to apply different colourings. 

 
31:46 
So if I want to switch to looking at a hydrophobicity, hydrophilicity view, I can do that right in platform. 

 
31:50 
It recolours all the monomers that way. 

 
31:52 
So we've built this sequence analytics tool in platform, and it understands that these entities, these biologic candidates are peptides, knows what they are and knows the sequences is associated with each one and can draw those connections for you. 

 
32:06 
So you can analyse sequence and platform. 

 
32:09 
So I'm going to pull away from that for a second because again, we're going to start building up the data that we need and to build the to get a view onto that data that we need to be able to triage these candidates. 

 
32:19 
As long as you have integrated with your different sources of information, whether it’s your database, your repositories, we can give you access to views on this data. 

 
32:28 
So if you look at this, what I'm opening right now is our data in columns tree. 

 
32:31 
It gives you access to be able to add different information to your live report. 

 
32:36 
Right now I'm going to add some experimental assay data. 

 
32:38 
So I'm going, you know, 4 down here and I'm looking at my experimental assays and I want to focus on my wet lab assays right now. 

 
32:45 
So I can go ahead and any asset that I've conducted that's hooked in here, I can take that data and you can either unfolder it and see there's multiple properties. 

 
32:54 
Do I just want to add one or do I just want to add everything about that assay? 

 
32:57 
So I'm going to go ahead and just add that here. 

 
33:01 
As you add data, it automatically adds from the data in columns tree. 

 
33:04 
I want to have a view on to that. 

 
33:06 
It automatically adds that into your live report. 

 
33:10 
And some of these things like it gives you automatically defaulting to giving you, let's say the geometric mean for this value. 

 
33:17 
If you want to see the individual data points, I'm just selecting, you know, which view do I want to see? 

 
33:24 
Do I want to see unaggregated median mean mode? 

 
33:26 
I can go ahead and do that right in platform. 

 
33:29 
So if I were to go ahead and do that for all of my experimental assays, let's say I want to look at all of the data points and I want to, you know, add multiple assays, you can imagine that you would get to a live report that looks like this. 

 
33:41 
Here I have, you know, reporting of my IC50 Kd and Ki for all of these peptides. 

 
33:47 
What you'll notice if you look at these data points is you have multiple data points per entity and there's intracell alignment. 

 
33:52 
So each data point has all of the information for these three pieces of data aligned for each data point. 

 
34:00 
So you know that this first IC50 is for when the peptide was screened against MDMX, the next one's for MDMX, the next one's for MDMX. 

 
34:09 
OK, that's because I have 3 replicates. 

 
34:12 
I did it in triplicate and I have three IC50 values associated with them. 

 
34:16 
Same thing, you know, for MDM2. 

 
34:18 
You can see I screened it three times against MDM2 and I have data associated with that. 

 
34:22 
So you have all of these multiple data points for each one. 

 
34:26 
Like you can imagine having replicates or screening against multiple targets, you can have all of that there. 

 
34:30 
Now you can imagine this list could get pretty long if you're doing at multiple concentrations, even at more different targets that you're screening against, you could have data point after data point after data point. 

 
34:40 
It's hard when you're looking at all of this data, you know, in a grid like this. 

 
34:44 
How do I actually see, did it perform all against MDMX? 

 
34:46 
Did it perform all against MDM2? 

 
34:48 
Did it perform all against both but more one than the other? 

 
34:51 
And it's a little unbalanced. 

 
34:52 
So there are ways in platforms to do analysis where you can slice and dice the data points. 

 
34:59 
So what I'm going to do here is I'm going to look at the IC50 values. 

 
35:03 
I want to be able to slice and dice it based on the target that I screened against. 

 
35:07 
I want to see, did it do well against MDMX? 

 
35:09 
Did it do well against MDM2? 

 
35:10 
Because when I look at all these numbers in a grid, each of these data points all on top of each other, it's really hard for me to discern that I want to make this as easy as possible. 

 
35:18 
Again, the platform is all about making the analysis as efficient as you can, and you want to get a good view on to what you're trying to see. 

 
35:27 
So I can go in here and I can define limiting conditions, which basically says I want to parse out a set of data points based on certain characteristics. 

 
35:36 
So the characteristic I'm going to parse it out based on now is the target that I'm screening against. 

 
35:41 
I only want to see the data against MDM2 because I want to get a good sense, did it perform well there? 

 
35:47 
I'm going to call this IC50 MDM2, and I'm going to hit create. 

 
35:56 
And what you're going to see is that right next to that assay value, all those six assay values, I now have the IC50 for MDM 2. 

 
36:05 
And it automatically gives you the geometric mean of that so that you can look at on average. 

 
36:09 
How did it perform? 

 
36:10 
Well, did it actually do the right thing? 

 
36:12 
Do you trust it or not? 

 
36:14 
I'll try to build trust here. I'm going to go to the on aggregated mode and you'll see that it knew to pick out these three data points because the metadata tag on that data point, as you see from the column to prior to that is MDM2. 

 
36:27 
That's what it was screened against. 

 
36:28 
It knows that it picked out the right data points. 

 
36:31 
And so now you can get a view on how it actually performed against that specific target. 

 
36:37 
You can imagine doing that again for MDMX. 

 
36:39 
You have multiple targets, or you want to pull, you screened at multiple concentrations. 

 
36:43 
You can imagine applying similar conditions to tease that data part and get a real read on how it performed at different conditions, not just all the data layering on top of each other. 

 
36:52 
Now I'm a visual person. 

 
36:54 
So yes, it's great that I was able to pull out these numbers, but I just see a  sea of numbers that's really difficult for me to even look at and see good, bad or indifferent. 

 
37:03 
You can apply colouring rules to basically colour, I like to go with green is good, red not so good, just, because stop lights help me keep that in mind of what I'm actually trying to do. 

 
37:14 
So you can put colouring rules in platform to basically define different ranges of where you think it was good, I'd want this value to be below this number or above this number and to have that as a quick visual on your overlaid on to your actual numeric data. 

 
37:30 
So I'm just going to go ahead. 

 
37:31 
I will say anything below 10 is good. 

 
37:35 
I want to label that as green. 

 
37:37 
And then I'm going to add another rule and say anything above 10, let's start it at green because that point is still good. 

 
37:44 
And then to the maximum, I'm going to say anything that goes to that maximum is red. 

 
37:47 
So you're going to get this colour graduation, anything that's under 10, that's good 100%. 

 
37:52 
And then a read from green to red as it starts to get kind of iffy. 

 
37:56 
So I'm going to go ahead and apply the colouring rule. 

 
38:03 
Did I hit, I might have hit the wrong thing, Auto, auto. 

 
38:07 
Let me just make it more obvious. 

 
38:16 
There we go. 

 
38:18 
Some things were a little too close to that 10 and that graduation, as you saw, went from 10 to like 20,000. 

 
38:24 
So just to give a little bit more of scale on that, I just changed the colouring rule. 

 
38:28 
But did you see how quick it was to update it? 

 
38:30 
I didn't like that visual, didn't give me a good sense as I was moving away from the good values. 

 
38:34 
Just quickly went and updated it and now I can see what I want to see. 

 
38:37 
And again, you can layer many colouring rules, not one or two. 

 
38:40 
You could do 5-10, whatever you need to. 

 
38:42 
So now I've shown you how you could slice and dice data to take a whole slew of data points and pull out a piece of interest and multiple pieces of interest. 

 
38:50 
You can keep doing that again and again. 

 
38:51 
I've shown you how you can apply a colouring rule to be able to get a quick visual on that column. 

 
38:58 
What's good, what's in the middle, what's I don't even want to touch that one. 

 
39:03 
I'll point out to you. iU you want here, let's say that this is the parameter that I really care about. 

 
39:08 
I want to make sure that all of those really good ones are at the top. 

 
39:15 
I can set it to order my biologics. 

 
39:17 
I basically all I just did is I went and I put a sort and I did ascending. 

 
39:21 
So all the ones with super low values for my IC50, the lower the better are now ordered on top. 

 
39:26 
So I can focus on those. 

 
39:27 
And you'll see as you move down, things will gradually change. 

 
39:33 
Now we're going from yellowish to orangish and it'll keep going. 

 
39:37 
So you can apply filters to enable you to then sort based on certain parameters. 

 
39:43 
You also can look at this spreadsheet and say, OK, this is really, really busy. 

 
39:47 
I can't even tell where my IC50 data ends and my next one begins. 

 
39:53 
You can take a series and all I did was hit shift. 

 
39:55 
I selected all of these columns. 

 
39:57 
You can group them together to create yourself visual umbrellas. 

 
40:04 
OK, all my IC50 information is here. 

 
40:07 
Make sure that I kind of group that so I can see this is where I'm looking for IC50. 

 
40:10 
The next group would be Kd, the next group Ki. 

 
40:14 
And the last couple of things I want to point out to you, but they'll make a little bit more sense when I show them to you. 

 
40:19 
And the next sort of a fully analysed view, because I'm just giving you one offs of how you could do each of these steps. 

 
40:26 
But you imagine you want to do a lot of them over and over again and make a fully altered spreadsheet that has these different views onto the data. 

 
40:35 
So I'm going to show you in here like right now where we've been doing is we've been adding a lot of experimental assay data and kind of slicing and dicing that up. 

 
40:43 
But you can imagine there's other information you want to add here besides just pulling the data and having a view on the data. 

 
40:49 
If let's say you want to add a formula, I want to take, let's say my kD values and calculate delta G off of that. 

 
40:55 
You could go ahead and add a formula there. 

 
40:56 
And I'll show you that at the on the next view, just by hitting new and you'll come into this window where you can create a formula, create its name, put the expression and platform and add a formula to platform to do a calculation right there happening live. 

 
41:09 
And as you add more and more biologic candidates, it will automatically calculate the formula as it says, here's a new candidate, here's the data, here's the calculated result. 

 
41:19 
And I also mentioned to you multi parameter optimisations. 

 
41:21 
Again, the idea that you can point to several different characteristics and get it distilled down to one number of how well it performed. 

 
41:29 
You can again create one of those where you choose different properties to add to it. 

 
41:34 
Choose any of your properties and start adding you know those into layer into your multi parameter optimisation that will distil them all down to one number. 

 
41:42 
And I'll show you what one of those looks like on the next few and free form columns. 

 
41:47 
I just want to point out this is a really great commutative and collaborative tool. 

 
41:51 
The idea that you can add a place where you can communicate information, either free text very often if you want to, let's say, create a comments column where people can add I thought this, you know, molecule was really interesting. 

 
42:04 
Maybe you want to avoid this one. 

 
42:06 
Have that interdepartmental and even global communication between teams. 

 
42:11 
And also there's a few, you know, other types in there. 

 
42:13 
One of them that I really like is kind of the true false. 

 
42:16 
If you just want to have a check off.  Did this do well in this assay or not, you can create a free form column where people can check off this met the criteria. 

 
42:23 
Please, you know, take this candidate forward. 

 
42:26 
So this is just pointing out to some of the things that you can layer into the spreadsheet, some of the tools you can use to slice and dice and revisualize data to make it easier for you to digest. 

 
42:35 
And if I sat and I did that for all of the data and laid it out exactly how I want, we might be here for a while. 

 
42:40 
So I'm going to go to my next view, which is me taking everything that I just showed you. 

 
42:44 
Nothing. 

 
42:45 
Everything I showed you is in this and nothing. 

 
42:47 
I'm hiding and slicing and dicing up and visualising the data how makes sense to me. 

 
42:52 
So you can see I have my IC50 data, Kd and Ki. 

 
42:57 
It's been sliced and diced better based on whether or not it was screened against MDM 2 versus MDMX. 

 
43:03 
So I have my numeric values for each. 

 
43:06 
Those means I have colouring rules applied so it's really easy for me to see visually. 

 
43:10 
Oh, this one did great against MDMX. 

 
43:12 
Not so well against MDM2. 

 
43:14 
That might not be the best for that. 

 
43:16 
And what you'll also see is we talked about formulas. 

 
43:19 
So under Kd I have those formulas layered in to calculate my delta Gs right in platform. 

 
43:24 
So if I go to edit formula, you can see the formula that I put in, I just added that it's there. 

 
43:29 
Anytime a new candidate gets added, it has the data, it will automatically calculate that. 

 
43:36 
And here's a nod to that MPO again, I have a binding kinetics MPO. 

 
43:40 
Basically it's pointing to all of the Kd Ki IC50 data and distilling it down to one number. 

 
43:47 
The redder that is. 

 
43:48 
So the closer it is to 0, the worse it performed overall. 

 
43:52 
The greener it is, the better performed overall for all those characteristics. 

 
43:55 
But if you hover over any of these data points, you can see how it actually performed for each of the characteristics that I fed into the MPO. 

 
44:03 
So IC50 against MDM2 and MDMX, Kd against both, Ki against both. 

 
44:09 
So you can see, oh, what really weighed this down. 

 
44:12 
I was really the KD and Ki screenings did not go well here. 

 
44:15 
Actually did OK again when I just measured IC50. 

 
44:20 
And this right here is showing you some free form columns that I added a peptide affinity pass as people are looking through here and there's deciding this one is absolutely no good. 

 
44:31 
They can just come in here and X that off. 

 
44:35 
They can say in a free form column. 

 
44:38 
This candidate is not ideal because and my typing is getting worse as I'm going through the sentence and have that communicated. 

 
44:48 
And again, this is not your own personal spreadsheet. 

 
44:51 
I don't know if this was said clearly enough. 

 
44:53 
This is a web-based informatics platform that could be accessed by anybody who has access to this platform and this project. 

 
45:00 
So if you have project team members, whether they're in like me and they're in New York, they're Nas, they're over here, we can all be looking at this same exact view onto the data. 

 
45:11 
We can be communicating live. 

 
45:13 
So if you look here right now, actually there is somebody who is looking at this. 

 
45:17 
Let's see if I hover over, it's the demo user. 

 
45:19 
But if you had other users here, see this little person here with a one, let's say there were five, siz or seven other team members looking at this and trying to analyse and collaborate simultaneously. 

 
45:29 
You would see under this little person all of their names listed, and they could all be checking off adding comments and working together live. 

 
45:38 
And I think that's really something I want to drive home. 

 
45:39 
This is a live interactive commutative platform. 

 
45:43 
So we've shown you all these views onto the, you know, onto the data, how you can make things analytically accessible to your eyes by slicing and dicing, colouring rules, adding formulas, all of these wonderful tools. 

 
45:54 
But in the end, we're still in the spreadsheet view. 

 
45:57 
Againbeing a visual person, I want to look at data more visually, more analytically with plots and different spreads of the data. 

 
46:08 
There aren't just numbers in a grid. 

 
46:10 
So I'm going to show you a forms view that I created to analyse this group of peptides. 

 
46:14 
And again, this is customizable. 

 
46:15 
I created this layout. 

 
46:16 
Your layout can look very different. 

 
46:19 
So here we're evaluating the group of peptides. 

 
46:22 
I did keep my spreadsheet view in the upper left because again, I'm collaborating with colleagues that might want to have, you know, that spreadsheet look on the data and be able to actually look at the raw numbers. 

 
46:31 
But down here I have three plots and what you'll notice is only the first one is populated. 

 
46:35 
I have set this up to be a drill down where one cascades into the other. 

 
46:39 
I have to select things in the first plot and what I want to pass forward to the next one. 

 
46:43 
So my first analysis here is looking at the IC50 values, MDM2 versus MDMX. 

 
46:48 
So you have MDM2 IC50s for these candidates on the Y MDMX on the X. 

 
46:53 
I want those IC50s to be as low as possible. 

 
46:56 
So what I'm going to do is just select, we added these horizontal and vertical cut off lines to communicate to the team. 

 
47:04 
We want things that are below this number. 

 
47:07 
So I'm going to select all of the candidates. 

 
47:10 
Sorry, my track pad slipped a little bit. 

 
47:14 
Let's say all of the candidates in that section of the plot and what you saw do it again is automatically my KD plot to the right of that populated. 

 
47:23 
So anything that I felt past the criteria for IC50 is now in my Kd plot. 

 
47:28 
MDM2 versus MDMX again, want to make sure I take all those bottom ones. 

 
47:34 
And because Kd and Ki very often are similar, happens to all be the same. 

 
47:40 
What you'll notice is as soon as I do that three-layer triaging by those 3 characteristics is that all of the candidates now populate. 

 
47:48 
They've drilled down into this table at the right-hand side. 

 
47:51 
So you could see all the candidates that performed well against all of those properties here. 

 
47:55 
But also my, I have my sequence viewer in this analysis forms view because in case I want to take a quick look at the sequence of these molecules and see is there anything I want to pass or fail? 

 
48:06 
Or do I see, does it have like a weird proline or there's a cysteine in there. 

 
48:10 
How did I miss that? 

 
48:12 
You have the sequence accessible right in front of your eyes at the same time as you did this data drill down. 

 
48:18 
So now what I can say is, OK, everything looks good here. 

 
48:22 
They passed all the criteria. 

 
48:24 
I'm going to go ahead and check all of these guys off and I'm not going to be too critical right now. 

 
48:29 
I'll just check all of them. 

 
48:31 
I'm going to pass them for peptide affinity. 

 
48:33 
Their affinity was good across the platform. 

 
48:35 
I didn't see any red flags, so now I've checked them off saying I want to move them to the next stage of the process. 

 
48:41 
Again, this is about a workflow, how you can move candidates down as you funnel down from a large group to your finalists. 

 
48:47 
But how did that communication actually get to the next team member so that they see that that's happening? 

 
48:52 
Well, let's say there's a next screening stage of the process. 

 
48:54 
They can have their own live report, which I'm going to click on right now to the right of that. 

 
48:59 
And this live report, you see it was empty, and it just automatically populated. 

 
49:03 
The reason it did so is what I've done, and I'm going to go up here and just show you because I've set up a search. 

 
49:08 
So anytime the affinity pass is true, it automatically populates. 

 
49:14 
So if I go back here now, right now, you saw there was 13 in there. 

 
49:19 
If I go here and then let's say I decide to go back to spreadsheet view again, you can go back between these tools. 

 
49:27 
Just going to check off a couple of more. 

 
49:28 
I'll check off that guy. 

 
49:31 
I'll check off that guy. 

 
49:33 
So I added three more. 

 
49:34 
So now if you go back to the next live report, instead of 13, it now has 16. 

 
49:40 
So you can set this up to automatically populate this next live report, this next view onto the data to only take things that have passed a certain stage. 

 
49:49 
So if you're ready, let's say the person looking at this live report is the next assay step. 

 
49:53 
They are automatically seeing when a new candidate enters their workflow step. 

 
49:59 
There's no lag time. 

 
50:00 
The candidate automatically enters their queue. 

 
50:03 
They know what they need to do next. 

 
50:05 
And to make it even easier, there is, I'm trying to find it here, but it might be because I'm in demo mode. 

 
50:14 
There should be under here a subscribe to notifications button. 

 
50:17 
Basically what that is, is let's say this person isn't looking at this live report every day to see three new candidates got added. 

 
50:27 
You can subscribe to notifications where you automatically get emailed that you have new stuff to work on. 

 
50:33 
This is the next big thing we need to get this out. 

 
50:36 
So now we're at the point where showed you LiveDesign, how it's been, you know, initially adapted for biologics, some of those embedded tools, how to get your data in, how to look at it in analytically accessible ways, both through a live report and those, forms views, where you build out, you know, your plot and how you want to sort of stage things and how you can pass things to the next step. 

 
50:55 
You can imagine doing this over and over again for each step of the process to make it seamless. 

 
51:00 
At this point, we're going to take a little coffee break that has a lot of information to digest. 

 
51:06 
When we come back, we're going to talk about all the fabulous predictive tools you can use to add and enrich these processes. 

 
51:13 
And then I get to come back again and show you how you can use all of that information in this platform as well. 

 
51:17 
It's not, oh, you do these predictions. 

 
51:19 
Oh that's totally disconnected. 

 
51:21 
No, the idea is to marry these worlds and really make this impactful. 

 
51:25 
We're going to kind of shift gears. 

 
51:26 
We've spent the first part really talking a lot about LiveDesign, how to use it for, you know, for small molecules, how it's being transitioned for biologics and how to use it in a triage and workflow. 

 
51:37 
At least with these initial steps being very focused around experimental data. 

 
51:41 
Now we're going to start going to the other side. 

 
51:44 
I'm really talking about some powerful predictive tools that we have at Schrodinger and eventually I get to come back again and show you how we can use those as well in LiveDesign and get everything under one hood. 

 
51:54 
So I'm going to introduce my colleague Dan Cannon, who's going to walk you through the modelling approaches. 

 
52:10 
So I'm Dan Cannon, I'm an application scientist at Schrodinger and my role is to act as a scientific consultant so that our customers, our clients can really make the most from our software to really add value to their modelling to their discovery pipelines. 

 
52:28 
So Schrodinger Suite, we have offerings for a great many things and I'm not going to go through all of these today, but you know, from structure prediction, from structure refinement analysis, protein engineering, some of which I'll show today. Protein interaction, molecular dynamics, there's a really a huge amount of technology that is accessible to you through the suite. 

 
52:51 
Some of the key things that are relevant to what we're going to show today are mutational analysis methods. 

 
52:57 
So there are two kinds of main ways to approach mutagenesis in silico. 

 
53:05 
MMGBSA often referred to as residue scanning. 

 
53:08 
This is where we basically take, our amino acid side chain and quickly swap it out into a little bit of refinement. 

 
53:16 
This is extremely cheap to run, so it takes like less than a minute on your laptop. 

 
53:22 
And of course, what it allows you to do is ideate certain things faster and come up with a score effectively. 

 
53:30 
Is this a good mutation? 

 
53:31 
Is this a bad mutation? 

 
53:33 
We make certain approximations here. 

 
53:35 
So it's not rigorous delta delta Gs or Kd values that we get from this but I would say it's more of  a ranking. 

 
53:41 
These are better mutations than these. 

 
53:44 
For more scientifically rigorous calculations, we use a technology called FEP+, which we'll talk about more in a moment, but we have fewer approximations here. 

 
53:55 
We have explicit solving. 

 
53:57 
We do dynamics, we do effectively enhance sampling as well, so that when we make a certain mutation that it's not just trying to fit in. The mutation that you're making is not accommodating to the surroundings, but it's mutual. 

 
54:14 
Yeah, it's a kind of induced fit effect so that you can actually predict that. 

 
54:18 
And with these you can actually predict more closely true free energies. 

 
54:22 
And this has been, the reason that it actually works so well is that we have put in quite a lot of, scientific development into this replica exchanges. 

 
54:34 
This REST technique is what we use to allow the local protein environment to reach our sample more quickly. 

 
54:45 
Co-Alchemical ions, so if we're making charge changing mutations, one thing that we need to do is actually balance the charge and we do that in a clever way during FEP. 

 
54:53 
GCMC solvent is something that we do. 

 
54:55 
This is grand canonical Monte Carlo. 

 
54:56 
Basically what it means is that we can effectively solvate hard to solvate pockets. 

 
55:01 
So if you have a very tight pocket, there may eventually be some water in there, but to allow water to permeate in there is a bit difficult. 

 
55:09 
And of course our force field, our mathematical description of all the system and FEP has been shown. 

 
55:16 
There are several publications looking at this for small molecule FEP, neutral mutations, charge changing mutations. 

 
55:22 
One thing I didn't include here is actually on homology models for protein stability as well. 

 
55:27 
Works very well, protein stability. 

 
55:29 
And also we can apply this to protein residue pKas as well. 

 
55:32 
So this is something I won't present right now, but please come and talk to me about this. 

 
55:38 
And so just to give you a little idea of what it looks like, this FEP is we're effectively switching off the wild type and switching on the mutant. 

 
55:47 
And as it does that, it allows the system to move dynamically and reach equilibrium much more quickly. 

 
55:57 
We can apply this for stability when we're comparing our folded state with our unfolded state. To do these horizontal transitions is very difficult. 

 
56:08 
So what we do is we make those vertical transitions, which is what you saw the switching off of the wild type and switching on of the mutant. 

 
56:20 
And what we have in front of us isa few maybe common data sets you may be aware of the Pucci and the Mayo data set, and these are characterised protein mutations and their effect on stability of the protein. 

 
56:34 
And these cover, you know, polar charged all these different types of mutations. 

 
56:39 
And the bottom line is that it performs very well. 

 
56:42 
What you see here is we can't model that full unfolded spaghetti. 

 
56:47 
Therefore, what we do is we take a molecule as a surrogate, either the monopeptide, so just amino acid or tripeptide, the mutation site plus and minus one. 

 
56:57 
And when we do that, the errors quickly decrease, meaning that basically, as you can see in these correlation plots here, is that we get a good correlation with things like thermal stability and thermodynamic stability. 

 
57:12 
We can also apply this to affinity.