0:00
So I would like to introduce John Shelley.
0:03
John earned a Master's from University of Waterloo in Theoretical Chemistry and a PhD from University of Pennsylvania in Computational Chemistry.
0:10
Following a post doctoral research in computational chemistry at the University of British Columbia, he worked for Procter and Gamble studying surfactant structures and solution.
0:21
For the last 24 years, John has worked for Schrodinger LLC as a scientific software developer and a research scientist, managing a number of products, including using the Material Science Force Brain product.
0:33
John has focused on computer modelling of drug formulation for much of the last eight years.
0:41
I, if at any time I wander away from the mic and you can hear me just shout and I'll try to correct it.
0:49
There's a lot of areas where computer modelling can be applied to drug formulation.
0:54
I'll kind of give a high level overview of that.
0:57
And then the main part of my talk will be about a particular application study for mRNA and liquid nanoparticles.
1:05
Now you probably most of you probably don't know much about Schrodinger or we're more in mostly in drug discovery.
1:12
So I'll talk a little bit about our background.
1:14
I'll also, I would say the type of modelling we do, molecular modelling is not heavily used in formulation.
1:20
So I'll try and give a perspective on the impact potential impact of molecular modelling on drug formulation in general.
1:28
And then I'll go over a little bit of our software platform and then go into the spectrum of modelling for formulation.
1:35
I'll go through three different case studies fairly quickly before getting to the lipid nanoparticle modelling.
1:42
And then finally talk a little bit about how you could potentially work with Schrodinger.
1:51
We interact with people in three different ways.
1:55
Our dominant way of generating revenue is through software licencing.
1:58
We have 1500 distinct customers in the world in various industries.
2:06
The largest one is pharmaceuticals and most of this is in drug discovery.
2:10
We're we use a style of modelling which seems eminently adaptable to drug formulation.
2:17
We also do contract research.
2:19
If there's a particular problem you want us to study, then we can do that.
2:24
Often this is a good way of prototyping modelling applications for your company rather than invest in someone yourself.
2:31
You can pay us, and it can be handled faster and cheaper.
2:35
And then finally, we engage in fairly intense collaborations with various customers to actually solve problems.
2:45
We have offices throughout the world, more than 800 employees.
2:50
The nice thing here is it means because we have offices of so many places that there's a good chance of a face to face meeting.
2:56
And also, if you write to us for support for your software, then someone in one of those time zones is probably going to answer you fairly quickly.
3:07
Value of modelling, I would say for drug formulation, it's a very interesting field for me because I think there's very good opportunities for modelling.
3:14
And the challenge in my opinion, is that in many drug formulation and delivery problems is that there's an organisation just above the molecular level.
3:25
And that's not very easy to characterise experimentally.
3:30
You can tell, so sort of aggregates or inhomogeneities and for instance, dissolving drugs, those are very hard.
3:36
You can tell they're there, you can tell how big they are, but you can't tell much about the internal structure.
3:41
And modelling is kind of the mirror image of this is we're very good at telling you the internal structure of these things, but we're not very good at telling you whether they will be there or not or how big they'll be.
3:51
So, and I think the reason why it's important from the experimental side is that these aggregates break the relationship between molecular properties and product performance.
4:01
There's something complicated happening between those, and molecular modelling can help fill in that picture.
4:07
The other thing eventually with modelling can do is start making quantitative predictions.
4:12
A lot of the areas in drug formulation, we're not quite at that stage.
4:19
OK, software platform: So, most of what we do is structure based modelling.
4:22
By that I mean we work with three-dimensional structures for individual molecules, or we build large collections, some molecules and calculate the interactions between those molecules and the calculations are physics based.
4:36
How do these molecules interact with each other or what are the inherent properties of these molecules?
4:42
Another area which we're expanding into, and we have a group dedicated to is machine learning.
4:47
We do the whole gambit of the sort of very old school machine learning QSAR all the way up to deep learning techniques.
4:56
And we also have a tool for collaborative informatics.
5:00
And what this is, it's a web-based tool where you can have the data, experimental and computational data in a spreadsheet that everyone in a meeting can be looking at the same time, even if they're in different cities or different countries, which we do fairly often.
5:14
And then also with different skill sets.
5:17
So medicinal chemists in the same meeting with computational chemists in the same meeting with formulators and allows you to make decisions based upon the data collectively rather than, you know, one particular group make calling the shots.
5:30
And that's how we actually do our drug discovery programmes.
5:35
OK, on the physics based modelling, we have techniques ranging from very computationally intensive but also quite accurate, through a series of approximations up to methods which are more approximate but can deal with larger systems at much longer time scale.
5:56
So it ranges from traditional quantum mechanics on individual molecules, perhaps in an implicit environment.
6:03
This here basically represents quantum mechanics on molecules in an explicit mole environment.
6:10
This is a very common technique, atomistic genetics.
6:13
So atoms are spheres, you can connect them together in a way which resembles molecules, and you can carefully parameterize these models to make them reflect reality and to be able to bring predictions.
6:25
And so we're actually making quantitative predictions of drug-protein binding using these technologies.
6:31
Now if you want it and for nanoproducts you need to go bigger.
6:36
You can go up to a level where it’s very similar type of modelling, but the spheres now represent functional groups rather than individual atoms.
6:48
We have experienced in a lot of areas that are relevant for drug formulation.
6:53
I'll only be talking about ones in bold here.
6:56
So crystal structure prediction, API encapsulation by cyclodextrin, excipient selection for ASD formulation and then LNP nanoparticles, cyclodextrin is used.
7:09
You know, we have a lot of small molecular drugs that are not soluble enough.
7:13
So cyclodextrin is used to help enhance the solubility so we can do simulations where the complexes form, we can add these cyclodextrins and derivatives tend to cluster in solution that clustering may be important.
7:26
We can also study that clustering.
7:29
Using different techniques, we can also get at the geometries of the inclusion complexes and the binding free energies of the inclusion complexes.
7:37
This is a collaboration we're doing with AbbVie.
7:40
It's entering, Samuel Kyeremateng is here.
7:43
He's from the collaboration.
7:45
He's probably entering about 8 years now.
7:47
We've been working together.
7:49
So Samuel, Matthias Degenhardt, Kristen Lemkemper and Ekaterina Sobich are specialists in amorphous solid dispersion delivery of small molecule drugs.
8:01
And they've encountered some surprising problems.
8:04
And because it's a platform for them, instead of just working their way around the problems, they try to understand them.
8:11
And they had invested a lot in understanding a particular type of problem.
8:16
And after they had a hypothesis, they engaged us to try and see if we could support or counter the hypothesis.
8:25
And I won't go into it in the interest of time in too much detail, but it's kind of more of the problems illustrated here.
8:33
So they simplified it down to very simple amorphous solid dispersions, a binary one of two polymers Copovidone, CPV or Soluplus, SLP in combination with ibuprofen, kind of a hydrogen atom of drug molecule studies.
8:48
And the problems they encountered were most severe when they want to formulate the drug and it's, it has a carboxylic acid in it, and they want to formulate it in the protonated form.
8:58
And for copovidone that really slows down significantly.
9:03
For Soluplus, even though the polymers are fairly similar, it basically shuts down.
9:08
There's no drug release, which is surprising.
9:11
So their hypothesis, I won't be able to go into too much, is that there's a key residue in or key monomer in each of these polymers.
9:20
So in Soluplus it's vinylcaprolactam and in Copovidone it’s vinylpyrrolidone, they're very similar.
9:27
The vinylcaprolactam has two extra CH2 groups in it, and their experimentation suggested that the drug is more effectively hydrogen bonding to the vinylcaprolactam than it is the vinylpyrrolidone, thus locking the solid form in place.
9:42
And then also they find conversely that the water hydrogen bonds more effectively to Copovidone than it does to Soluplus.
9:50
So there's a weaker driving force for water to enter into Soluplus.
9:54
And they didn't tell us their hypothesis until after we did the study.
9:57
And then we went back and looked, we found the same pattern of association between those functionalities.
10:03
We also found that in our simulations, we do see the slowdown for both of them and the more dramatic slowdown for Soluplus.
10:11
And we also see a structuring that happens on the surface of the [colon] Soluplus, which we believe is also a factor.
10:17
And so that's maybe an added bonus that you get from doing the simulations.
10:22
OK, we also are doing crystal structure prediction both as a service and as a software offering.
10:30
And the reason here is for small molecule drugs, many of them are delivered in the crystalline form.
10:35
You need to have very good understanding of your crystalline forms when you're delivering small because you can get in trouble if you don't have good control of it in production or storage.
10:46
OK, now on to the title case study, I won't go into this too much.
10:52
We have some background here, but I think there's some, you know, LNPs, RNA, cationic ionizable lipids, there's two states, protonated or neutral, cholesterol helper lipids, in this case DSPC and a pegylated lipid.
11:08
And we're going to be imitating the Pfizer BioNTech vaccines in these studies.
11:14
One of the key areas of challenge is the bioavailability of the nucleic acid polymers, even though they're successful, you know, the actual effectiveness is maybe a percent or something actually is translated.
11:30
And so we're thinking of bringing modelling into this area and we have engaged in collaborations to aid in understanding of the barriers for delivery process.
11:44
And one of the key things here is, OK, if you're looking at varying composition of your lipid nanoparticles, then it's a really good example of what I said is you get this supramolecular organisation, and the organisation matters for the effectiveness of the therapeutic.
12:02
And it's hard to characterise.
12:03
There are very good experimental studies, but they tend to end up doing some sort of spherical averaging of the particle, which makes it really hard to get insight into the particle structure.
12:13
And we also think that again, modelling can help give a much clearer picture of that and also under understand the barriers to effectiveness.
12:22
We'll be studying 2 sizes of lipid nanoparticles.
12:26
The, this is the full size one and it performs well.
12:30
It's just really slow to model this.
12:33
If you can decrease it by almost a factor 2, the calculations are about 30 times faster.
12:40
And so we do both.
12:41
We find actually the results are very similar for each, and our technology isn't great for telling you size distribution.
12:48
So this isn't something we believe it.
12:52
So here we do simulations of self-assembly in three stages.
12:56
The first stage is in the acidic ethanol water mixture, and we put in the ionizable lipid is 90% ionised, 10% neutral.
13:08
And we start off with random confirmations or random systems for each the efficient and the full size.
13:15
And here's the simulation of methanol.
13:18
It forms a single lipid nanoparticle in this simulation; it ends up forming a few here it's about the same size.
13:25
These lipid nanoparticles are larger and more dynamic than you're used to seeing in the end product, which is, you know, in an aqueous solution, it's the ethanol pumping it out.
13:37
OK next stage in our process then is to replace the ethanol.
13:44
We can do this computationally, we just magically replace the ethanol with more water and continue the simulation.
13:51
And what we see is, you know, initially there's a lot of ionizable lips out in solution because ethanol was there.
13:57
Those rapidly condensed on the lipid nanoparticle.
14:00
Lipid panel particle becomes smaller, and it becomes a little bit more spherical and less dynamic and something similar in the full size one.
14:10
We actually have run this out longer because it's taking longer to evolve.
14:13
That's one of the penalties of doing bigger things.
14:16
We're at about 15 microseconds now in that one and it is solely agglomerating into a bigger lipid nanoparticle.
14:25
Here's what these look like.
14:27
So this is just without the showing the solvent.
14:30
This is looking from the nanoparsec in the outside.
14:32
The blue is the ionised lipid, the red is the RNA.
14:36
There's a cross section.
14:37
And if you look at some of the earlier talks had sort of schematic diagrams of what this looks like, those schematic diagrams, but it came out of our spontaneously came out of our model.
14:48
We didn't build this in.
14:51
And then finally, here's a cross section with the water actually visualising, you can see there's quite a bit of water inside.
14:57
So our models or the simulation gives the mRNA buried inside the water channels inside the lipid nanoparticle.
15:05
Some water channels lack RNA, which is also observed experimentally.
15:09
We get about 25% water inside the, I think experimentally it's down around 20.
15:15
It's a bit of noise, but it's pretty good PEG lipids on the outside and then little or no neutral ALC is surface exposed.
15:25
And these are all in agreement with what's known from experiment.
15:29
We can also do density profiles of the lipid nanoparticles with respect to the centre of the lipid nanoparticles.
15:35
So this is water density, and you can see it levels off inside.
15:40
This is the possibly charged ionizable lipid.
15:43
This is cholesterol, the black line, and the red line is the RNA.
15:47
We can also look at density profiles relative to the backbone of the RNA.
15:52
So what is interacting directly with the RNA, and it tends to be predominantly the positively charged ionizable lipid water has a distinct high density right beside it.
16:04
No surprises there.
16:05
But this is coming out of our model spontaneously.
16:09
And then the other components are a little bit of cholesterol nearby. Lots of things structurally that we're getting correct spontaneously out of this self-assembled lipid nanoparticle, which suggests that we're we should be able to be predictive as you change the composition of the system and want to see what is effective if you can self-assemble it.
16:33
And presumably as you change composition, you can see how the structure changes.
16:38
And the next step, the third step in this is that, you know, lipid nanoparticles are created under acidic conditions, but there's, you know, around pH 5 roughly, but there's tend to be stored under blood pH condition 7.4.
16:53
And so you have to raise up the pH and we do that in steps.
16:57
So we just take our, some of our possibly charged ionizable lipids and convert them into neutral ones and then continue the simulation for a while and then repeat that process.
17:07
So up here we, this is as they form and as we're converting to the green neutral ionizable lipids, you see the RNA which is red tends to separate out and coming down here, it separates out and remains within the lipid nanoparticle down to even fairly low levels of ionisation.
17:26
And this is of course a well-known phenomenon, lipid nanoparticles, it's blebs and at least some groups associate, Peter Cullis's in particular associate these blebs with better RNA stability inside the lipid nanoparticles.
17:45
And so we're getting this form, we didn't train it for this.
17:48
It's kind of like we did a study, we self-assembled and said, OK, what do we do with it afterwards?
17:52
OK, let's change the pH effectively and see what happens.
17:55
And it spontaneously did the right thing, which is really encouraging.
18:02
So what's the next step?
18:03
So we're starting to look at a major bottleneck in the delivery process.
18:08
And so the lipid nanoparticles, they undergo endocytosis as the endosomes is evolving, the RNA has to get out of the endosome into the cytoplasm in order to get translated.
18:20
And the effectiveness in the vaccines that are on the market is about two or three percent of the RNA that's inside the endosome actually gets out into cytoplasm and translated.
18:31
And so this endosomal escape is a key aspect of this process.
18:35
And so we've done, we've started doing a few prototypical studies of endosomal escape so that we can start trying to understand why the escape is so limited and what's, what changes to the formulation, how they affect that.
18:51
And so this is just one example we've done, oh, probably about 20.
18:56
So I don't different simulations like this.
18:58
I don't think we have enough to say statistically reliably, but it's, I think it's a very promising technology.
19:04
So in this particular simulation, here's the lipid nanoparticle.
19:08
We position it right beside a model membrane for the cytoplasm.
19:13
It's negatively charged.
19:14
It's under acidic conditions.
19:15
So the lipid nanoparticle is positively charged.
19:18
So they should want to interact, and this is very similar to other people have done.
19:22
But our model is very efficient and fast.
19:27
And so we can actually see the RNA coming out, a fairly long RNA coming out as the simulation proceeds.
19:35
It takes about 6.5 microseconds before it completely detaches on cytoplasm side of the membrane.
19:42
So this is encouraging.
19:43
We can actually simulate the release.
19:47
We can also simulate the failure of release too, here's the same sort of set up, a different one.
19:53
We have happened to have two RNAs and this one's orange and one's cyan.
19:58
And as the simulation proceeds, the cyan one decides it's going to come out on the endosomal side.
20:03
And this is spontaneous.
20:04
We did nothing to make this happen.
20:06
And So what it's showing that from our simulations is that spontaneously we can see both.
20:14
You know, successful release and unsuccessful release.
20:17
And so the next stages are now to study, hopefully with a partner, because we're geeks, to study the actual process as a function of composition and see if we can help people understand the barriers trained as a new release.
20:33
And so I think this takes us back to where we started.
20:36
You know, formulators are really good at what they do.
20:39
They almost always succeed.
20:41
They do so systematically, but because I think there's this sort of intermediate levels of structuring, they have to do a lot of experiments.
20:48
And I think modelling can help inform the formulators about what's happening.
20:52
And so not that the modellers will tell them what to do, but the formulators will be able to think more effectively about their systems.
20:59
And so I think simulation will help there if it at least in the near term, should hopefully lead to more efficient development of drugs.
21:10
And I think in principle, as I think formulators understand their systems better, they will be more actually more, they will be able to derive more optimal outcomes with the help of modelling.
21:21
And so just in summary, there's a very broad range of formulation calculations that we've been doing.
21:32
They almost always have to be in collaboration with industry because we need the feedback on what's happening and its relevance.
21:41
We, I talked a little bit about cyclodextrin complexation, amorphous solid dispersion, dissolution and crystal structure prediction as kind of diverse applications.
21:51
And then most of this we talked about lipid nanoparticles, and I think our technology can really help fill in the gaps of the knowledge and how things change as you change the formulation.
22:03
We're actually engaged in some very neat the studies that we just started two weeks ago.
22:08
These things take a month or two or three months to run.
22:12
So, and hopefully in two or three months we'll have some very interesting stuff on some new formulations that people have been talking about.
22:20
And I think the main thing I would say here is we can also do things like if you're loading targeting ligands and you're wondering what stage of the formation process do you want to load them or are they really exposed on the surface of lipid nanoparticles.
22:36
Those are things that we can also help with that I haven't talked about in principle.
22:41
We can also help with the passive targeting or you're changing the lipid composition.
22:47
We have no work in that though that I can talk tell you about.
22:51
But I think I'm very excited about being able to see the RNA release both successful and unsuccessful.
22:58
And I think it's very promising that we can help move the field ahead.
23:03
And with that, you know we are, we're at booth 22.
23:06
We have a good sized team here actually we have a part of the team is actually for drug discovery.
23:11
That's actually our main market.
23:14
But we also have a good representation from drug formulation, and we can help with all aspects and a lot of things in formulations.
23:21
People actually don't have a lot of computational research, so how do they grow into it?
23:25
And we can help a lot with that, both providing software, we can evaluate technology to see if you really want to get into it without investing a lot.
23:33
We can provide set up access on cloud computing.
23:37
We can create workflows if you'd want to set up your own cluster, we can help with at least the computational chemistry side of setting up your cluster, paid research and collaboration.
23:47
So we have pretty much the whole spectrum.
23:50
So I'd love to meet with any of you or my colleagues who would also love to meet with any of you now or after, thank you.