0:49
Yes, thank you for the introduction.
0:53
So today I just want to spend some time telling you about what we're doing at Twist Bioscience, how we're kind of unlocking new avenues in antibody discovery with our innovative tools and our technologies both in vivo and in vitro, and how we're really taking an AI/ML approach in discovery as well.
1:13
So I'll give you a really brief introduction on Twist, who we are, what we do, what our general capabilities are.
1:19
Then I want to move on to a case study on where we've used an in silico approach to discovery.
1:27
And then I want to give you some of the newer tools that we have just an indication of what we've got there to really help with antibody optimization and in particular affinity maturation.
1:41
So to start just a brief introduction.
1:44
Well, we're a DNA synthesis company, we make DNA, we're using phosphoramidite based chemistry to make DNA which has really been miniaturised down onto silicon platform roughly the same sort of footprint as a 96 well plate.
1:57
But it allows us to make a million oligos with base by base precision on the silicon chip up to now 500 nucleotides in length and we're stretching that still.
2:09
We're trying to get up to 700 at the moment.
2:12
And we're making lots of DNA.
2:14
So we're making about 16,000 oligos a day in our factory in Wilsonville, Oregon. So The amounts of DNA we make is a lot.
2:26
And to be able to cope with that, we've built this state-of-the-art infrastructure.
2:30
So a lot of proprietary software, a lot of robotics and we also have this integrated e-commerce ordering and tracking system that's really beneficial.
2:41
So it's the combination of the technology and the logistics and infrastructure that allows us to provide, to develop this kind of game changing throughput and to provide products that are really cost effective, high quality and delivered at speed.
3:01
So talking about quality, it's the oligos that come off that silicon platform that have industry leading quality.
3:08
It's the silicon that allows us to control that chemistry really well.
3:13
We have 100% oligo representation, so no dropouts.
3:16
We have incredibly high uniformity as shown by that plot on the left.
3:20
So everything is within threefold of the mean.
3:22
And then we have a low error rate of one in every 3,000 bases.
3:26
And it's these oligos that we then use as the starting point to build a set of custom products.
3:34
So we can elute these single stranded oligos off the chip in various ways and just provide our customers with these high-quality oligo pools often for CRISPR screens.
3:45
That's very useful guide RNA, things like that.
3:48
We now can amplify those oligos, make them double stranded and provide you with double stranded oligo pools, right?
3:55
We can take these oligos and build them, assemble overlapping oligos together to make gene fragments, clone those into vector, your vector, our vector to make sequence perfect clonal genes.
4:08
So that's just some of the examples of the products that we can make.
4:12
But talking about clonal genes, we've been working really hard to drop the timelines there.
4:18
So again, it's a logistics thing.
4:21
We've got really good at it.
4:23
It's a really smooth system.
4:25
And now you can basically put your order in of sequences of inserts on a Monday and we'd be ready to ship out sequence perfect plasmids clonal genes in by the Thursday, Friday of that same week.
4:38
So within four business days.
4:42
The benefit to us of this express genes workflow is the products that require this internally at Twist.
4:50
One of those is antibody production.
4:54
So by having the express genes, we've been able to speed up our antibody production workflow where we can go from customers inputting sequences to us being ready to ship purified antibodies in as little as 10 business days now.
5:12
And just to show you a little bit more data on those turnaround times, this is data of the past year.
5:25
So orders can be in the size of handfuls of antibodies or hundreds of antibodies or sometimes even 10 plates worth of antibodies in an order.
5:35
You can see that we're consistently maintaining this turnaround time of 10 to 18 business days.
5:42
This is for our one mil transfections and if we just looked at the cell lines for HEK293 transfections, we're going from sequence to purified protein in 10 to 15 business days and a slightly longer for CHO, so 13 to 18 business days.
5:58
But the aim of showing you this is just to show the consistency of the turnaround time.
6:04
So it's not just antibody production that we do.
6:09
In terms of supporting those in the antibody discovery space and the antibody optimization space, we do have a whole suite of industry leading tools and technologies all under one roof.
6:24
And that means you can come to us with your target of interest, your wish list, your TPP, and we can generate a strategy internally, at Twist and take that discovery campaign all the way through to provide you with development ready drug candidates.
6:39
That's one option, but we're also very flexible in how we work.
6:44
So if you just wanted certain tools for your own discovery, we can certainly provide that.
6:50
Or if you wanted to use our discovery capabilities up to halfway and let us provide you some sequence information, that's also a possibility.
7:01
So it's a very flexible system in terms of how you can use this.
7:06
So just looking at this in terms of repertoire generation, we're often making custom libraries for customers.
7:14
And it's just again, our synthesis platform that allows us to do this where by having base by base precision when we're building oligos, we can absolutely replicate designs that people have of naive discovery libraries, for example, and provide libraries of 10 billion variants in this space.
7:35
But alongside the custom libraries, we also have sitting on our shelves a bunch of discovery libraries that are ready to licence out for phase display.
7:46
And we often use those in our campaigns as well.
7:49
And then on the in Vivo side, we can immunise alpacas for VHH discovery, we can immunise rabbits, humanise mice and we have our own proprietary diversimab and divergimab mice which are hyper immune mice with a broken immune tolerance to allow for a broader epitopic space.
8:12
So that's repertoire generation and I'll tell you a little bit more about that in our case study where we're using slightly different approaches as well.
8:19
But in terms of what we do in selections and discovery is we have an in Vivo approach where you can use single B cell sorting with the beacon and a more traditional hybridoma approach as well.
8:34
And then with in vitro, we can do phage display or yeast display.
8:40
I mentioned our high throughput antibody production, but on the back end of that we're also doing high throughput characterization.
8:48
So this is for binding for kinetics affinity ranking in terms of having a number of Carteras on site for SPR, but also octet for BLI.
8:59
But more so now we're seeing a lot more requests for developability and we can certainly do developability assays in high throughput too.
9:09
And I'll show you some of the assays we can do.
9:13
And finally, if required, we can of course provide optimization as a service, but also provide you with tools such as libraries, our new products, Multiplex gene fragments, which I'll show you about in, I'll show you a bit about in a minute as ways of kind of optimising or affinity maturing or humanising or removing any kind of liability.
9:35
So many ways to work with us in the discovery space.
9:41
So I'm what I'm going to do is I'm going to move on to talk to you about a specific case study.
9:46
This is for a GPCR target C5A receptor one.
9:50
We wanted to pick something difficult here.
9:53
We wanted to pick something that's of importance in the immuno oncology space where we think actually there aren't enough candidates.
10:05
So we thought it's a good target to pick for this approach.
10:11
We used more of an in silica approach alongside doing an immunisation approach and then alongside doing our using our libraries that are kind of off the shelf in vitro discovery libraries that we can use for phage display.
10:27
We did this in silico approach.
10:29
So here from a structure of C5A receptor, we were able to use kind of neural networks to decipher what are druggable epitopes.
10:41
And then using those we were then able to run this antibody design algorithm to identify millions of candidate antibodies structures that can bind to these epitopes.
10:54
Then we used a neural network filtering step and reduce the numbers down to a manageable 10,000 sequences.
11:03
And if we look at those sequences, they were pretty unique.
11:06
So we noticed that we got over 6,000 unique heavy chain CDR3 sequences from there.
11:14
Now often we get asked this often that why don't we just take the 10,000 sequences through for checking for binding and functionality.
11:25
And what we wanted to do is really increase the chances of success here.
11:30
So from this de Novo starting point, we kind of see maybe a 1 in 1000 success rate for binding and we really wanted to increase that.
11:41
So what we did is we built a CDR shuffle library with those 10,000 sequences.
11:49
This was a single chain library for phage display.
11:53
The theoretical diversity of shuffling all the CDRs was incredibly high at 1 x 10 to the 25.
12:00
But what we ended up building was a representative sample of that at 1 x 10 to the 9.
12:07
Now a library project allows us to really print individual CDR sequences and build such shuffle libraries.
12:17
And the libraries are always kind of NGS quality checked.
12:21
We get a lot of information from the NGS QC and on the right, I've just shown an example of one of the QC readouts here.
12:31
So we're looking at CDR length distribution and we're comparing the designed lengths in blue compared to what was observed from our NGS reads and it does match up quite nicely.
12:45
So this library was then used in phage displays.
12:50
So we carried out five rounds of cell-based panning and in the later rounds we carried out NGS sequencing.
12:57
And using the output from the NGS sequencing, we were able to identify a set of antibodies to reformat and we used an enrichment algorithm as well as an unsupervised clustering algorithm to select those 96.
13:13
Now for other campaigns, we've also taken it a step further and carried out a deep learning neural network algorithm to identify these low abundance, high affinity clones that you can really see over there.
13:30
So not sure that quite worked.
13:34
Yeah, right.
13:36
It's just on the top left, those crosses right up showing you really good affinity but really low abundance.
13:44
So the neural network method is actually very popular as well.
13:51
But for this campaign we use clustering and enrichment, and we were able to use our antibody production facility to, you know, make these as IgGs.
14:01
You can see that from the de Novo method, we don't get good expression for all clones.
14:07
Some don't really, but we got some really good high expressing antibodies from this as well.
14:15
So quite a range.
14:17
These were then taken through a set of characterization assays.
14:20
Now as it was a GPCR target, the first thing we wanted to do was check cell binding.
14:26
So we identified as a hit anything that showed a threefold improvement in cell binding over the parental.
14:36
And then we titrated those to get kind of putative EC50 values.
14:41
And we were actually able to carry out some SPR on this where we got our GPCR within my cells and we're able to use that on SPR to get some kind of kinetic information there.
14:54
But for this particular project, the most important thing we were looking at was functionality.
14:58
So we ran a reduction in beta-arrestin recruitment.
15:02
So we wanted to look for a reduction in beta arrestin recruitment.
15:05
And you can see that we did identify two or three clones there that did show a reduction in Beta arrestin recruitment, but they weren't quite as good as the clinical control of avdoralimab in the dark blue line there.
15:19
So we did have to go back and optimise this.
15:23
But other things we're doing.
15:24
So we've run this sort of strategy multiple times Now.
15:27
And for this case study, we didn't really do any further characterization.
15:31
But for others, we're very much moving in to carry a characterise in terms of developability.
15:37
So we're running analytical SEC on thousands of variants.
15:41
We're running hydrophobic interaction as a column based method, but also in a more high throughput way where we can take thousands of samples at a go in a day.
15:52
We've got the nanotemper to run DSF and DLS there, AC-SINS and also BVP-ELISA.
15:58
So we can do a full package of developability assays on hundreds or thousands of antibodies in parallel.
16:08
So for this anti the leads that we got off this campaign, we did have to go in and use our own proprietary method for generating libraries that we then reran and in phage display campaigns this time both cell based planning and micelle based planning.
16:27
And we were then able to identify functional leads which had greater potency.
16:34
So some even better than the avdoralimab control.
16:40
So we managed to get a 16-fold improvement in potency there.
16:44
So regarding affinity maturation and antibody optimization, there are a number of ways in which Twist can help.
16:54
Often people come to us with their lead sequence wanting us to help with designing a library that they want to then screen internally themselves.
17:05
That's possible.
17:06
We can certainly help taking, you know, lead sequences and building precision libraries where you only move a certain distance away from each CDR just to stay close to the lead sequence.
17:20
We can build CDR shuffle libraries.
17:22
We can build libraries where you may have some detail on structure and know exactly what positions to diversify.
17:30
Again, the libraries are like highly uniform and can be built to high diversity so that can be shipped to you to use.
17:41
If you require humanisation, we have these in silica humanisation methods that allows us to really very quickly spit out humanised antibody sequences and then using our genes workflow and our high throughput antibody production facility, we can very quickly make these into purified antibodies for testing.
17:59
And then although we have our proprietary method for affinity maturation and antibody optimization using a library-based approach at Twist, what we're also seeing more recently is AI/ML based affinity maturation where actually we're having lists of variable region sequences where those need to be built exactly as is as a library.
18:25
So these defined variant libraries of entire variable regions.
18:30
And that's now something that we're very much able to do.
18:35
If you recall, I said we're now printing oligos of up to 500 nucleotides in length and that easily covers a variable region.
18:44
And that means instead of building libraries where we shuffle CDRs, what we can now do is print entire VH regions or VL regions or VHH regions where you can keep the trios of CDRs, you can keep any changes in the frameworks and just have exact defined variant libraries there.
19:07
So we did this actually for not for optimization, but it was with the Baker lab where they had de Novo design a set of VHH antibodies using ProteinMPNN and then fine-tuned using RoseTTAFold2.
19:25
And they built these sets of 9,000 VHH antibody designs to four different targets.
19:34
So they didn't want any combinations.
19:36
They wanted pools of exact 9,000 VHH sequences and we were able to build that with our Multiplex gene fragments, which they then ran on yeast display and showed that they were able to identify binders to all four targets.
19:53
So that's where this new tool, the multiplex gene fragments is coming in very strong.
19:59
We're trying to push that length, as I said to 700 and 750, 800 hopefully to cover single chain FVs.
20:08
But at the moment this is what we've got and it's quite a useful product in AI/ML.
20:17
Just to summarise, there are a number of ways in which you can work with us and you can use our kind of tools, technologies, our services for your discovery campaigns, for your optimization campaigns.
20:30
So do come and speak to us if anything resonates or if you want any more information.
20:35
But just to say, although Twist is a DNA synthesis company and we're making a lot of the products that others are offering, some of the ways in which we differentiate ourselves is kind of listed on the left hand side.
20:50
It really is quality.
20:52
It really is just being sustainable by having this miniaturised process.
20:56
It's a green process.
20:57
We're quite proud of that fact.
20:59
We can deliver exceptionally fast and with scale, with lots of numbers, with good amounts.
21:05
So it's an affordable kind of set of solutions and one of the kind of key differences is we're made-up of a team of scientists who have expertise in various areas that you can absolutely dip into and use.
21:20
So I'm going to stop there.
21:20
Thank you so much for listening and if I have time for questions, I'm quite happy to take some.
