European Society of Human Genetics Conference (ESHG)

*This transcript was generated using an AI-based transcription tool and may contain errors. It reflects the spoken content of a live webinar and has been lightly edited for clarity.

Thank you, Charlotte. And really, a very, very warm welcome from our side, from the Roche Sequencing Systems side. You are choosing to spend the afternoon with us, and over the next one hour or so, we'll talk through some of the concepts on the AXELIOS platform and also hear from some of our early evaluators. 

Before we go to the evaluators, I'm going to give you an introduction to the platform. Some of you are already familiar with it, with the platform quite well, and for some of you, it might be a little bit more new information. We also have the platform in our booth, 440. So if you would like to take a look there, there's a lot of team there that are also looking forward to speaking with you. 

AXELIOS 1 platform is the first of its generation in the SBX technology. It is a two-instrument platform, and I'll just give a brief introduction. On the left-hand side is our synthesis instrument that accepts library pools. So you make libraries the usual way that you do for other platforms, and then those libraries are loaded onto the synthesis instrument. That instrument then converts the libraries into what we call synthetic molecules called Xpandomers, and these Xpandomers are about 50 times larger than the original DNA template. And that makes the sequencing measurement much simpler and robust during the downstream sequencing process. 

The sequencing instrument is also in two parts. The sequencing module itself has a nanopore-based CMOS sensor module that is reusable, unlike a traditional flow cell. So the Xpandomer molecules, as they pass, the base calling is done in real time. That is done through the analysis module, which does not only the base calling but also post-primary processing, and we'll talk a little bit more about that as well. 

The whole orchestration of the order of the samples from the library all the way to analysis and the data flow happens through the AXELIOS 1 platform manager, which is also a single touchpoint for users to set up user management, user permissions, etc., in the lab. And that manager also performs some additional functions. We'll talk a little bit more about that as well. 

This platform is really built based on the SBX technology, which has a cornerstone around scalability and flexibility. That is something that you have heard from us multiple times before. You'll actually hear it from me over the next 15 minutes and also from our evaluators. From a throughput perspective, it's a really high-throughput platform, but it is also very flexible and scalable based on the throughput needs of a lab. 

For example, in duplex mode-and we'll talk a little bit more if you're not already familiar with what the duplex mode is- in just a 4-hour sequencing run, 16 genomes can be sequenced. So the whole concept is shorter, faster, flexible batches. But if you had a larger sample size or a high-throughput facility, you can actually do those 4-hour runs in a consecutive fashion, so that allows you to scale up to 64 genomes in one go as a full run queue. And during that 4-hour run, per 4-hour run, you're generating approximately 1.8 to 2.7 terabases of concordant duplex bases. And that throughput is sufficient for more than 30x coverage across all of those 16 genomes in that 4 hours. 

And then there is another mode of sequencing with simplex, where you can generate about 40 billion total reads in a 4-hour run, depending upon the sample and the library conditions. So for something like a 170 base pair counting application, 40 billion total reads would be the expected throughput, but if your read mode-reads were-reads can be adjusted based on your target insert size, and we'll talk a little bit more about that as well. 

And that really gives a lot of flexibility to the platform, from the architecture of the platform with its two-instrument system, as well as the ability to do both the duplex and the simplex mode, which is a highly flexible method of sequencing, really allows a lot of different methods and applications to be implemented on on the platform. We'll talk some of that about with our evaluators, as well as the last presentation in which we'll talk about some additional simplex use cases. 

In addition to that, various sample types can be easily implemented as well. We'll show some of that data for genomic DNA from high molecular weight genomic DNA as well as some other sample types. And our third speaker today will actually present a very interesting sample type with filter cards. 

The modular system architecture, with the synthesis instrument separated from the sequencing instrument, also provides a lot of flexibility. The synthesis instrument, for example, can process four library pools in one go, and it takes only 4 hours. And then those library pools can be independently loaded onto the sequencing system. But this type of flexible architecture means labs can actually run their workflow based on their sample requirements versus-versus based on the rigidity of the architecture of the platform. 

And in the simplex mode, the system is also capable of processing longer reads. We've actually shown quite a few data on this in the various conferences last year as well as AGBT earlier this year. Longer read lengths up to 1500 base pair, which is possible by appropriate sample and library prep conditions, of course, and then we have shown the distribution from 200 base pair to 1500 base pair, and you can just adjust the median insert size on the longer end of that spectrum based again on your specific sample or experimental needs. 

In the simplex mode, you get accuracy which is greater than Q20, and that raw read base accuracy is sufficient for multiple applications, for example, in counting. But for applications where you need higher accuracy, the duplex mode is really the workhorse for things like whole genome sequencing where it enables more than Q38 base accuracy. 

As I mentioned, the sensor module on the sequencing instrument is reusable, unlike the flow cells that you use on other systems. It can be reused up to 20 times, and based on that, as well as some other efficiencies that are built into the platform and the flexibility of the technology, it really offers a very competitive operational expense at various scales and various throughput. 

We have talked about the platform pricing as well as OpEx pricing in various forums. I'm not going to repeat it today, but if you are interested in getting more understanding of your pricing or in your specific country, please do come by our booth. We have representation from various countries in Europe as well as from the US, and if you're interested in understanding more of that pricing, we'll be happy to discuss it with you. 

The last piece I wanted to talk about from a feature perspective is scalability. And I mentioned various flexibilities that are already built into the platform, but one of the things with-and like the 16 genomes in a 4-hour run, but the ability to set up four runs on the sequencer back-to-back fashion to really have the capability to scale up. In addition to that, on the synthesis instrument, you can go from one to four library pools simultaneously. So there's a lot of flexibility and adjustability for various throughput in a lab. The other is that these-the synthesis instrument and the sequencing instrument are not one-to-one hardwired to each other, which means one synthesis instrument can actually support multiple sequencers. You can also have a combination of synthesis instruments and sequencing instruments. 

So for example, everyone would start with one synthesis instrument and one sequencer, but because you can process four Xpandomers in just one 4-hour run on the synthesis instrument, you can have an additional sequencer if you wanted to prioritize for turnaround time. You could also have two synthesis instruments and eight sequencers and have your operational setup that way. So it really depends upon the throughput requirement of your lab, the sample cadence, and how you perform your library prep or store your library prep, because those library pools can then be queued up on the synthesis instrument, and those Xpandomers can be queued up on the sequencing platform. So lots of flexibility and scalability that is built into the platform. 

So with that AXELIOS 1 platform and the features, and the question is, okay, what does it enable? What type of workflows can we do? What methods are possible? So before we talk about that, just a general workflow. The workflow is not very different from how you would do sequencing in your lab, starting from library prep and all the way into sequencing and analysis. So I'll focus on the key differences that actually are driven from that flexibility that I was just talking about. On the library prep side, we are introducing AXELIOS Library Prep, specifically for duplex sequencing where there is a two-step ligation for a hairpin adapter, and I'll show you the structure in a bit. 

We include linear amplification in place of PCR amplification. And that actually gives several advantages in terms of the variant calling efficiency, and we've shown that data before-how you get really good small variant calling as well as copy number and other efficiencies that come from linear amplification. That linear amplification process, the time of that process depends upon your starting input. In general, when we do a duplex whole genome sequencing workflow with 50 nanograms of input, the linear amplification time recommended is about 6 hours. So that makes your library prep approximately 11 hours of total turnaround time. And once you do quantification and pooling, you put it on Xpandomer synthesis. As I mentioned, it takes 4 hours for that run. And on the sequencing system, it takes 4 hours of sequencing time and 1 hour of setup time for a 5-hour time period. And then you can set up all those four runs consecutively for a total runtime of 20 hours. 

So when you put all of that together, and you started with, let's say, 16 whole genomes with 50 nanograms of input each, you can actually complete the whole 16 genomes in less than 24 hours of turnaround time, or 64 genomes in 36 hours. And that is very much faster than the typical runtime that is enabled on another sequencing platform. So the whole process is quite fast. And if you then think about library prep and the input requirements and the amplification time, then based on your amplification time, your workflow can actually go even faster, as we demonstrated with SBX-Fast in its world record. And I will actually not talk about that. I'll have the next speaker, Sean Hofherr from the Broad, talk about that, as well as the new amplification-free method that we have been working towards. 

This library prep is actually very straightforward again. It's built on our KAPA technology, and it can be easily automated on multiple platforms. And I'll actually speak about that. The other difference I wanted to mention is that the runtime that I'm mentioning on the-on the far right on the-on sequencing, that actually already includes your downstream analysis all the way up to BAM creation. So there are features built into that compute that comes with the sequencer that allows flexibility in the terms of the files that you can generate, but it already includes the primary processing as well as post-processing. We'll talk about that as well. 

These are the two run modes, as I mentioned briefly earlier, duplex and simplex. So the top one is the duplex run mode where both parent strands are physically linked with a hairpin adapter that completes that contains the sample index, and you basically run through the read from one strand to the other and drive intramolecular consensus for high accuracy greater than Q38, and also resulting in very high accuracy and a small variant calling performance, and Sean will show some-show some data on that. 

On the simplex side, there are certain applications that do not require very high accuracy, but you do need a lot of reads. And for for things like counting applications in RNA or proteomics, etc., simplex is with its Q20 base accuracy a very powerful technology, a very flexible technology that allows you to create not only a lot of reads like 40 billion reads in 4 hours, that can also have a capability to go much longer reads up to 1500 base pairs, as I mentioned. And there are other ways, as we have shown previously in target enrichment, for example, you can also have UMIs and drive intermolecular consensus to again leverage those large numbers of reads for other applications that may require certain other considerations. 

On the AXELIOS 1 Library Prep, again, it's built on KAPA technology, very straightforward. Linear amplification is very straightforward as well. We have demonstrated the-um-automation on multiple different platforms. The kits are just built for automation. And I want to briefly show you an experiment that we ran in-house, starting with that 50 nanograms of workflow, as I mentioned before, which is our standard for SBX duplex. Combined 16-plex libraries together, ran it manually so that we could do a comparison with the automated methods, and ran it on three different platforms: Beckman Coulter, Hamilton, as well as our own AVENIO Edge platform, and then did a library quantification and pooling downstream, and then ran our 4 hours of sequencing process. 

AVENIO Edge specifically can do quantification and pooling on-deck as well, so a complete walkaway system. And so this is just an example to say that in each of those cases, there was sufficient yield that was achieved from all the automation systems, and in fact, the CVs are quite tight for those yields, specifically on the AVENIO Edge platform. We actually see that in our other KAPA library preps as well where that platform does give you very tight CVs across the yield, and these yields are sufficient for multiple runs on the sequencer. And then we have done the downstream analysis on each of those libraries as well. I'll show you one example of comparison of the performance for manual versus AVENIO Edge across coverage, the duplex bases, the insert sizes, as well as the Phred score for quality, and we see these are actually very consistent with each other. And I could literally replace AVENIO Edge with a Hamilton or a Beckman, and the graph doesn't really change. So the performance across each of those platforms is quite similar, and we're very happy to see that not only these platforms but also newer platforms, we will have AXELIOS 1 Library Prep automated. So the app for the Volta Callisto system, for example, is currently in development and will be available later in the summer as well. 

So with that, and we'll cover some of these methods today, and the final speaker, Patrick, will also talk about some of the simplex methods. But just in the context, this is an open system. Every method and application is possible. We have already demonstrated quite a few of them in the past year and a half. I will have a QR code at the end of the presentation, and that allows you to just review some of the presentations we have done over the last year and a half across a multitude of these methods. And at launch, many of these protocols will be very readily available for whole genome sequencing across different sample types, as well as SBX-Fast and single cell RNA. This is a collaboration that we have with 10x Genomics. And then for the remainder of the year, we will continue to release more app notes and data sets. A data set for whole genome sequencing is already available on our website. Variant calling is also available as well. And then from there, we'll continue to build additional protocols and expand really on the various methylation, target enrichment, as well as proteomics methods coming soon, and we will continue to build from there. 

Before I conclude my presentation, I wanted to talk a little bit about analysis strategy as well. So our analysis strategy is based on two components. One is the integrated analysis that is available on the local compute, as I mentioned, a compute module that comes with-with part of the sequencer, and that takes care of a lot of compute-heavy steps. So base calling as well as just generating all the way up to BAM files, those steps can be integrated and completed pretty much at the same time as your sequencing is completing. So when the sequencing run finishes, you can actually have a BAM file. And then the standalone tools, which are primarily for variant calling, and I'll talk about that. But then there are also pipelines available for demultiplexing as well as generating those alignment files if your requirements were something different and those files were not already available as part of the integrated box. 

So there are three ways by which you can then have an analysis, an open format analysis file that is ready for downstream sequencing, and I'll talk in the next slides about each of those, starting from the first one with raw reads. Both for simplex and duplex mode, you can generate-basically your base call files and then take them off from the-from the sequencer and take it for standalone analysis. This is the most upstream point in SBX data where you can have access to that data. It gives a lot of flexibility for very advanced users or customization of application and analysis. 

For many users, for SBX duplex specifically, this next egress point will be more efficient where you can have demultiplexing as well as driving that intramolecular consensus for the high quality Q38 reads, again FASTQ format available from the instrument, and that can then go into your analysis downstream on your own infrastructure. 

And then for even more users, the most efficient way would actually be to leverage the built-in mapping and alignment tools on that compute platform and basically have a BAM or a CRAM file that is then ready for analysis in your standalone environment. So this reduces the data footprint as well as any cost that may be related to those mapping and alignment. So just as a summary, these are the various types in which you can actually generate the data off of the instrument, and we'll talk a little bit about VCF and the plans to actually enable that for SBX-Fast as well. 

In addition to XOOS tools, which I'll talk about in the next slide, there are various tools that are already built by the community. Now to be fair, they have been built not for SBX but for the other technologies that are out there, specifically sequencing by synthesis. Some of those tools are already compatible with SBX data type with minor or without any modification, which are listed here on the left-hand side. And some of the other tools are compatible with some optimization, and there will be documentation that will help us guide you through all of that. 

Now SBX data type is different. There's lots of resources available on our website that actually give you information on the data type specifically, so I would encourage you to check that out as well, and that can give you an understanding of how to use these tools if you were-if you were to use them. There are some tools that just ingest the data very differently, so we do not currently recommend to use them with SBX. 

But very importantly, on the XOOS side, the team has been really hard at work with some SBX-compatible and like really optimized tools. Some of these tools are already available on GitHub as open source, and many of those will be posted very soon. So you can read the list yourself, but suffice to say that not only do we have a lot of these fully optimized open software tools available on GitHub right from the very start, but we'll continue to build on those tools, including several biomarker and specialized callers in order to enable your research. 

My last section, I'll conclude by saying that we do not say that we can do this all by ourselves. There's lots of methods out there, both in terms of analysis tools as well as library prep tools. And this being an open platform, that's been our strategy, just working with the ecosystem both on the analysis side with NVIDIA as well as Google DeepVariant that we have talked about before. In fact, Google DeepVariant is already optimized for SBX at this time. But also our previously announced collaboration with 10x Genomics on single cell and spatial, and really very pleased to say that we are also now working with Olink to release SBX-compatible kits, specifically for the Olink Explore HT assay. They talked about that in their corporate workshop just a few minutes ago as well, as well as native support of SBX in the NPX Map, Insight, and Analysis. 

So we're very excited about that. But as I mentioned before, our strategy is to continue to work with other vendors in addition to creating, really leveraging the KAPA technology that we have as well as the XOOS tools that we are building out. This is my last slide. The QR code will allow you to go to our website that has a lot of technology overview, publications, lots of recordings there, presentations, data sets, especially the SBX data set that we launched before. And then we'll be here for questions after the talk. Thank you very much for your attention.

*This transcript was generated using an AI-based transcription tool and may contain errors. It reflects the spoken content of a live webinar and has been lightly edited for clarity.

Thanks, thanks for the intro. Thanks also to Roche for inviting me to give this talk. As Mitu explained, I'm going to be talking about a different protocol that we've been working with the Roche team to develop, um, kind of a modification that allows us to run still really fast genomes, but doing it when we're thinking about scaling up. 

So, just some disclosures, so all the work that I'm going to be presenting was done at Broad Clinical Labs. So we've had an early access instrument in our laboratory for about a year and a half now, and so all the work that we've presented in the past has also been done in our laboratory.

What I'm going to be talking about is my opinions, and not Roche's, and also the platform is in development. So, it's not yet released, and so everything I'm talking about has been done on a development platform. 

So, thankfully, because Mitu gave a good demonstration, a good explanation of how the technology works, I don't really have to spend much time here. Um, except to say that when we first met with the Roche team about the instrument, as it was like still not being talked about anywhere, one of the exciting things that came out of it was the speed and the batch size. And so that was, that was like obvious right away from the beginning. Um, and because of this flexibility, the kind of native application that I thought would be a great use of this is actually the Neonatal Intensive Care Unit. 

So, in the NICU, um patients come in, very young children, very critically ill. Um, a lot of them need genetic testing. Actually, the majority should get genetic testing, if not all. Um, a very high rate of patients in the NICU are positive for genetic findings, somewhere in the 30 to 50% range depending on the study. And even sometimes a negative result is also, along with those positives, guiding the treatment and management of that patient. And so, pretty much every child in NICU benefits from genome sequencing, and rapid genome sequencing is really important. The faster you get a result, the better the patient is managed, and also the better it is for the healthcare system, and the physician, and family. 

So there's been a bunch of interest and studies across the US, and ex-US, outside, and everybody's really excited about the NICU and genomes. This is not a new concept. But what is a new concept is really taking SBX and AXELIOS and shrinking down the amount of time that it takes to sequence a genome. Um, and so, when we started doing this, it quickly became apparent that we probably could have done the fastest genome, and so actually as, you know, as early in September, we ended up getting the record, as Mitu explained, for fastest genome. Um, we sequenced a genome in less than 4 hours. It was a lot of work, um but the cool thing about this project, um so it was, it was published in New England Journal of Medicine, and the cool thing about this is we didn't actually just focus on doing the fastest genome, so like that's a a neat feather in the hat to to be able to say you have a world record. My kids were kind of impressed with that one, um more than anything else that I've told them that I do. Um, but the cool thing was we actually in in this publication, which came out, which was based on that same work and was kind of an extension of it, we were able to apply this super fast genome sequencing to a research study with Boston Children's and showed that same day and actually same shift genome sequencing in the NICU is possible, which is pretty amazing. 

Um, it's also, although it took a lot of work to get there, this is not the kind of thing that is one and done like we did it for the flash and and the excitement. Like, this is reproducible time and time again. It's being done typically by one person, the two main people that are doing this work in our laboratory,Michelle and Sophie right there in the picture in front of the early access boxes. Um, and you could just see like, as we were doing that work, um we're constantly getting times below 4 hours and also slightly above 4 hours.

So in this study that we did and was published, with Boston Children's, Monica Wojcik was the the PI and really she's a neonatologist, awesome collaborator, and you know, the whole concept was to show that we we could get the result to the NICU way faster than what is traditionally being done. So right now, at least in the US, when testing is ordered, you kind of have some options on how fast the testing can get done for a genome. You could get a traditional genome sequencing, which is a couple of week turnaround time. You could get a rapid genome sequencing, which is typically around 7 days, or you could get an ultra-rapid genome, which is somewhere in the 48 to 72 hours. Um, and that time is typically like from receipt of sample at accessioning. And so, what we showed was we could actually do it in one day including transport, especially when it's a close location. So Boston Children's is relatively close, depending on the time of day and traffic, to where Broad Clinical Labs is, so we're able to get samples couriered to us. 

10 cases were run in total, seven of them were published, but we had to kind of close out the publication before the study was completed. Um, and three of the 10 had positive relevant pathogenic findings, seven of them were negative. Um, all of them were all these results, there was clinical testing being done in parallel, and as that clinical testing was coming through at a later time point, that was being used to guide the management of the patient, and we kept matching it up with our results, and in every case it was concordant with what we found. 

So this is what the timing looked like across all the different samples. Um, we had in the study, we had run some benchmarking samples. We had a retrospective cohort of samples, and then this prospective 10 patient study that we did. And all of these are including hospital pickup, and so you could just see that this is like a single shift in the NICU. Patients are getting seen, samples are getting collected, sent to our laboratory, run in our laboratory, interpreted by our laboratory, and reports are generated. Once again, just to highlight, this was all done on a research basis, so not yet validated, although as soon as we have the instrument in hand, we hope to validate this. And there were, of the three positives, they were pretty interesting cases. 

So the first positive case was a kid with irretractable seizures, and what we found was this pathogenic variant in KCNQ2, which fit the phenotype and was later found to be also detected by ultra-rapid testing. Patient was already on medication, was on the appropriate medication, but if this was done in real time, this could have guided the patient to the medication that actually treated the seizures. 

Second positive case was unbalanced translocation, so we detected a terminal deletion and a terminal duplication, different chromosomes, 14q and 10q. Um, and this was a patient with multiple congenital anomalies, and what you see here is that we found the findings from the test, from the genome, they were confirmed later through a whole bunch of different technologies. 07:38 

The third case, which came after the publication, was a patient that had an elevated C5 on the acylcarnitine portion of newborn screening came, and they were suspected to have isovaleric acidemia. They went to the NICU, they had hyperammonemia already at presentation. Um, the sample came to our laboratory, we got the result back actually before the metabolic testing was completed, which is pretty cool. That tends to be a pretty fast process to get metabolic confirmatory testing done, and we were able to get genome results at the same time. 

And so, you know, this has a huge lot of lot of broad applications on on why this is important and, you know, it's reproducible, it's useful, you could get 1 to 2 runs with a single tech, you're able to get your outputs into tertiary analysis platforms, which is really important if you're going to be generating any kinds of reports. And so really this technology kind of changes the paradigm for what could be done with whole genome sequencing, especially with same day. 

Um, and so, just looking at the workflow, um this is based on the SBX Fast workflow that Mitu described in her slide. Um, and you could just see that one to three genomes is around 6 hours, so it is an amplification-free workflow, um but that means we need a lot of DNA. So, the way I think of this protocol is like um you have a car that is a fast car and you want to get that fast car to be like as fast as possible and you start pulling off all of these pieces that are not absolutely critical to get it as light as possible. So it's kind of like what we did here, and we basically got a protocol to be as tight as possible. So one to three samples in less than 6 hours. Pretty cool. 

But, you know, and then you look at the other workflow, so this is the SBX-D workflow that Mitu described, which actually has the amplification. So the main key difference is that for this amp for this linear amplification workflow, 50 nanograms of input, so that's a lot less, so that's great, especially if you have like FFPE or some like limited sample type like the next speaker will be describing. Um, but that's not, that's not really something that we're typically struggling with to get 50 nanograms. And so, when we, you know, this is this is a fast protocol like Mitu described, so less than 24 hours for an Xpandomer with 16 genomes around 36 hours for four sets of those Xpandomers, so around 64 genomes in 36 hours, it's fast. But, we were kind of thinking is there some like middle ground where we get the benefits of SBX Fast, but we but we also get the benefits of SBX-D, and we kind of thought about what are places that we could that we could cut. And so we worked a lot with the Roche development team, um the team in Seattle put a huge amount of effort into really like getting this process to work and then transferred it over to us. And the new protocol that I'm going to be showing some data on is basically this SBX-D No Amp. So it's still the duplex sequencing, um just like the Fast and the normal SBX-D, but it doesn't have that linear amplification step, um it's been tuned for this workflow. 

And so now we're able to get 16 genomes in around 13 hours, um and four Xpandomers or 64 genomes could be done in around 28 hours. Um, 500 nanograms was the input, which is great, so that's that's actually kind of in the sweet spot of uh, you know, it tends not to be an issue to get 500 nanograms as input, and it's well worth it to shave off those hours compared to SBX-D. 

This is what the workflow looks like. I'm not going to go into specifics here except to say, um you know, that the workflow is relatively easy, it's been automated, and and yeah, I mean it works well. Uh, looking at the the three protocols kind of stacked up against each other, so um SBX Fast, 2 micrograms, 1 to three genomes, around 6 hours, um SBX-D, 50 nanograms, here you're talking about 16 or 64 genomes depending on if it's one Xpandomer or four, um and you're looking at 24 hours or 36 for the four Xpandomers. And then the SBX-D No Amp is actually this kind of like middle ground that we've been working on, which gets us um with 500 nanograms per sample of DNA, we get um the same number of genomes as SBX-D, so 16 genomes for one Xpandomer, 64 for four Xpandomers, and we're sequencing in 13 hours and 28 hours, respectively. 

So what we did to look at the data and how we set this up, we did some benchmarking with HG001 through 7, compared against high-confidence truth sets. Um, we then ran some cell lines that are have known pathogenic variants, and then we took those three positive cases from the the study that we did with BCH and in their NICU, and we ran it using SBX-D No Amp and put it back through tertiary analysis. 

This is what the quality looks like, which is very, very solid and comparable to what Mitu showed. Um, so right around 38 as our Q-scores. Um, we were focused on a 30x mean coverage, and so we did really well in the first two Xpandomer pools. We got, this was a this was a little bit ambitious and we threw in some sub-optimal samples, saliva, buccal swabs, and so, you see some of the coverage, mean coverage goes down a little bit in pool C or Xpandomer C and D, and that's really just because of some samples that, we're getting some bacterial sequence when we're getting human sequence as well, so it cuts down on the amount of human genome. 

Uh, when we look at the benchmarking HG001 through 7, it looks great. It looks very similar to what we've seen in the past, you know, across all different variant types, so the top panel is SNVs, the bottom panel is indels, you're looking at precision and recall for each of them, and it's separated based on different variant type within the genome. And the assay performs well. So we found similar results when we were looking at, SBX Fast, and so it was nice to see that it was also reproducible in SBX-D No Amp. 

With the the cell line, um similarly, so performance was as expected. So we tested a trio, this is a family, Coriell cell line family, where there's a proband that has a homozygous mutation in PCCA, so propionic acidemia, inborn error of metabolism. Both parents are carriers. Um, this is one of the parents as a carrier, heterozygous. Here's the other parent, same variant, also heterozygous, and then the proband is homozygous. 

So, exactly as expected, um for a single base deletion. Um, we were able to detect copy number variants, so this is a duplication in the DMD gene responsible for Duchenne muscular dystrophy. Um, we were able to pick up this 28 and a half kb duplication without any issue. 

Um, looking at a larger copy number events, so this is uh DiJorge, VCF, a 22q11.2 deletion, so it's quite large, 1.4 megabases. Um, and we were able to see it and detect it using the assay without any issues. 

Uh, and then last, this is a deletion, an insertion deletion, in CFTR, and we're able to detect this variant as well. Um, so as expected, the assay works well, so that's kind of really, you know, what we're striving for both on the benchmarking side and on the ability to detect pathogenic variants. Um, but when we are typically looking at cases, we’re not really like pulling up IGV to go variant by variant because that would be pretty impossible, so we use tertiary analysis tools. Um, the tertiary analysis tool that we're using at Broad Clinical Labs and has been validated for all of our clinical workflows is Franklin from QIAGEN. 

Um, and so I basically went in, ran three of the samples through Franklin that we had previously tested using SBX Fast, so this was part of that publication, and we just started from the DNA, went back through library prep, Xpandomer synthesis, and sequencing using the SBX-D No Amp, generated the data, got that data into the tertiary analysis tool, and we were also able to interpret and and identify and interpret the variants without any issue. So this is that patient that had the seizures with variant in KCNQ2. Um, this is the, similarly, we're able to find the patient with the two terminal chromosome deletion duplication that ended up being an unbalanced translocation confirmed later. And we're able to detect that and not have any issue there. 

Um, and then lastly, this was the case that was not a part of the study. This was the patient that had an abnormal newborn screen for uh isovaleryl carnitine elevation, and we're able to find the two variants in IVD that are both SNVs. Um, so yeah, with that I'll conclude. This work is, I'm kind of standing up here and presenting it on behalf of a lot of people that put a lot of time into this work. Um, Sophie and Michelle, I showed their picture, they're kind of the hands of all of this work. Um, and Edyta, who is like ferociously analyzing all the data as it's coming off in pretty much real time as fast as the instrument is generating it. Um, and then just the other collaborators at Broad, including Katie, who manages this project. Um, on the Roche side, we've had a huge amount of success in our relationship with the Roche team and it's just been great working with them. Mark and the development team in Seattle have they're kind of the they're doing a lot of the development work and then kind of handing it off to us based on what we're asking them to do. And so, it's a really good partnership between the two groups. And then obviously, we couldn't have done any of the work with the patients um without working with physician-scientists, and so the team of physician-scientists at Boston Children's are listed here, and then also the folks from Franklin, from QIAGEN, who helped us get all the data into Franklin. So, thank you.

*This transcript was generated using an AI-based transcription tool and may contain errors. It reflects the spoken content of a live webinar and has been lightly edited for clarity.

Thank you, thank you, for the introduction and for the invitation to share in the very preliminary and and recent findings of our collaboration inside with my disclosures. I would like to focus on the fact that CENTOGENE provided the samples, Roche performed the library preparation, sequencing, and data generation, and we performed the analysis of such data. 

So, just to give you a little bit of context of why we started this project and the reason for it let me tell you a little bit about CENTOGENE. So, we are a diagnostic lab with a global footprint, so we have we are actually this year celebrating our 20th anniversary um and our focus is in genetic diagnosis um clinical genetics for main medical genetic disorders, neurology, pediatric neurology, oncology, also now with a focus in reproductive health and preventive medicine. 

We have our database of over 1 million patients who help us here that also help us in the collaborations that we establish with pharma companies. So, we believe that on top of context, the regular, let's say, genetic diagnosis based on DNA other layers of multiomics will be and are already very helpful for the diagnosis of patients with genetic disorders. And we perform both the diagnostic test, so the DNA test, exome or genome sequencing as well as biochemical testing and RNA-seq on our filter card, we call CentoCard, for obvious reasons, right? Um but the CentoCard is essentially a filter card where we collect blood, and we extract DNA and RNA necessary. Also, we perform the bio-medical- the biochemical testing. 

So, now it's easy to understand that when we are considering or evaluating a new sequencing technology, it's very important to understand the performance for such a low input or for this specific use case because many of our tests are based on blood and most of our the samples that we receive come on this format. For G6 it works really really well. So, we really want to understand in particular how SBX performs with our, let's say, main source of samples. 

So, for that, we partnered with with Roche, with Roche and um we've selected for this first run for this first part of the proof of concept of this evaluation, 16 samples, three from blood collected in EDTA and the other 13 from the filter card. Of those, six samples had clinically relevant variants that were reported at the time of the initial test, the other 10 were negative. And the input for sequencing was 75 nanograms, and the run was um SBX duplex. So, run the duplex run for these 16 samples. As you can see, the six samples that had clinically relevant variants were all sequence variants only, different types. And I'll show you in a moment exactly what they or how they look like. Um but essentially, sequence variants only. These are the results of the QC from both um samples from filter cards and blood. So, in blood the blood samples are in red. In blue are the samples from the CentoCard. Um the number of reads- well, if you look across all the graphs, very, very similar profile, so no significant difference, which was exactly what we would like to have. And in fact, that has proven to be the case. 

So, um even if you look at the median coverage, all samples achieve 30x or more. The uniformity that is measured by the median over the 10th percentile coverage is also very similar for both blood and filter card samples. Same for the percentage of duplicates. The median insert insert length is also quite similar. It's around 220 base pairs. Same for the soft- the percentage of soft-based bases and the the quality score as well. I'd also like to highlight the homopolymer accuracy, both for stretches of eight base pairs or 10, but we achieved for all samples, for all the 16 samples, over 99.5% for homopolymer eight accuracy um score, and for HP10, 99.1% for all samples as as well. It's quite um let's say, a very nice surprise for us when looking at the data. 

So, I will go now and and show you how the variants that were called or how they look like. Um this is an IGV screenshot in gray are the the duplex bases and then in purple, the simplex tails. As you can see here, for example, this is a nice example of an RNU4-2 um so this is the common variant insertion of one base pair that was perfectly called. No issue. Um homozygous substitution, in this case, in APOE. Heterozygous substitution in TPM2. Now, one base pair deletion as well, here in ACE gene. Here you can see another substitution, heterozygous substitution. Uh another one as well. So, for all samples, for all the six samples, all the variant- the clinical relevant variants were detected and called. Here you can see a three base pair heterozygous deletion. In MYO18B. Here, a four base insertion. So, also clearly detected and called. A four base pair heterozygous deletion. 

So, very, very clean pileup and, so, nothing we can complain about, let's say. And the homozygous um substitution um in NGF gene and finally, a homozygous call, so homozygous substitution in DOK7. So, in summary, all the clinical relevant variants sequence variants were called, all with good quality, all in the- with the expected zygosity- in the expected zygosity. We achieved 30x or more coverage for all the samples starting with 75 nanograms. Um and the coverage the uniformity and the quality was quite quite good. 

So, as I said, these are very preliminary findings. We want to do more, of course, to complete this project and this evaluation. So, regarding these 16 samples, next steps will be, of course, to perform a genome-wide comparison not only of the the calls from from that are clinically relevant calls, but all around the sequence variants, copy number variants, repeat expansions, and others, so between our previous data set that was used for diagnostic workflow and and the the ones that we have now with SBX. 

And we also like to and we will analyze the 10 cases, the 10 negative cases that are from patients with genetic disorders to see if we can somehow increase the the diagnostic yield there, that some variant that might be interesting to then go and confirm in a clinical setting, that could provide, eventually, a diagnosis for for those patients. And then, we want to perform a second round, second set of of 16 samples also with more complex variants, including copy number variants, repeat expansions, and also variants in genes that are a little bit more hard to sequence, so located in more challenging regions um or more complex variants. 

Thank you very much to the CENTOGENE team that was involved, in particular Mandy. Uh that was here, I don't know where she is now. Um and to the Roche team for your support and collaboration, for this really very, very useful and productive. Thank you.

*This transcript was generated using an AI-based transcription tool and may contain errors. It reflects the spoken content of a live webinar and has been lightly edited for clarity.

Thank you, Charlotte, and thanks, everybody for sticking through to the end of this session. Four speakers is a lot, and there's a lot of other things, really interesting science, going on. And thanks to Sean and Jorge for the work they've been doing. It's really exciting to see AXELIOS being used. It's just really fantastic. 

I'm going to share two stories with you today, one of them focused on a novel application that we haven't shared before, and then another one on a novel use case where, thinking about how can you combine library types of different sources, different input materials, different species, and say, how could this happen in a real-life situation in a lab? 

So first, I want to talk about methylation and conversion chemistry. So we've not shown any data in the bisulfite sequencing space or the EM-seq, and so today I want to share with you an initial research protocol that's leading the way as we get forward towards launch to being compatible with workflows like this. We've generated two research workflows to work with existing libraries. One of them is a simplex workflow where we're able to take existing converted libraries or non-converted libraries, convert them to being compatible with AXELIOS, leaving this purple SBX handle on there where they can go into the Xpandomer synthesis. This is a simplex mode, so you're getting that Q20 quality. 

We've also generated a workflow where you can take a library and create a duplex-like molecule where you have a hairpin attaching a parent and daughter strand. What this allows for is the higher base calling at Q38 for that intramolecular consensus. But this isn't the same mode as in whole genome where we've presented duplex, because there you're linking parent strands together, you're going through the linear amplification, and you're maintaining that fidelity of the parent-parent. Here the duplex is really for intramolecular consensus for the Q38 base calling. As I mentioned on the simplex side, we're able to add on that SBX handle, which allows it to interact with the synthesis module. You go through Xpandomer synthesis and then sequencing, as Mitu and the others have described. 

So first, I want to present our bisulfite sequencing proof of concept workflow. For this, we used a contrived sample media where we designed oligos that were unmethylated, and we ordered oligos that were methylated. So it's a very controlled situation, so we could know what truth was. We performed the bisulfite conversion using the Zymo Research EZ DNA Methylation-Gold Kit. And then to-as a comparator, we used Illumina sequencing-based, and so this was the iSeq 100 with a 2 x 150 base pair using the i1 Reagent Kit. 

For the AXELIOS workflow, just as I described, we took these oligos, we treated them with the Zymo kit, we created the initial library, and then we performed the duplex conversion on these ones. And so we were looking for that higher base calling accuracy there. And then for the analysis, they both went through the same internal BWA-meth pipeline that we're using on-site. So what you can see is that AXELIOS is completely compatible with libraries that have gone through bisulfite treatment. 

So on the left-hand side, we're just showing a total conversion where we're looking at basically all CHs that exist. There's no methylation on these ones. And what you can see is that we're equivalent greater than 99% conversion in both settings, with very low false positives. So there's nothing preventing methylation conversion workflows from working with AXELIOS. When we look at on the other side where we we have protected bases, so methylated bases, you can see in our methylated controls, we're recovering the same 97-greater than 97% retention between both Illumina and AXELIOS. And so again, the conversion happens. Bases that are protected by methylation are present as well and able to detect that. And then our unmethylated controls in that same mixture look just like we showed on the other side where they are converted. And so what we're going to now move forward with is start using bisulfite and generating application notes for how to do this in a non-conversion workflow, so native workflow. And so this is really just to-to put out there that we can take converted libraries and sequence them directly on the AXELIOS. And then as Mitu shared, this will be coming later as an application in 2026. 

The next workflow I want to show is EM-seq. So here we took-working with a collaborator who will talk more in detail on the biology later on this year. But this was an initial pool of 16 samples that came from cell-free DNA. They also included pUC19 and lambda controls in there, so we had some methylation controls. And they went through a commercially available EM-seq kit. We took these libraries, which happen to be for another SBS system out there, and converted them into both the simplex mode as well as the duplex mode. So here we want to say, are we observing any differences between the simplex mode and the duplex mode in the ability to detect conversion? 

What you can see and I apologize for somewhat of a complex slide, but really, I just want to focus on the high level and the-the left and the right is that you're seeing very similar numbers and rates of conversion between both duplex and simplex. So in the duplex mode, looking at CpG islands in the cell-free DNA, you're seeing about 73%. And in the simplex mode, you're seeing about 73% converted. And then in the CHGs and the CHHs, you're seeing full conversion between both the simplex and the duplex mode. And then just really comparing this then on the controls where we know truth. Right, part of the problem with cell-free DNA and and doing these experiments is we don't actually know in all cases where the methylation occurs, especially here because these are an experimental set of research bank samples where we're expecting differences between them. And so if we compare the simplex and duplex on the lambda, very equivalent, with a slight decrease on the simplex. And then if we look at the pUC19 where we're expecting full conversion, we're seeing a slight increase. And this is really due to the difference in base quality, right? We're getting some more false positives in the simplex where it's Q20, whereas on the lambda methylation control, we're getting a few false negatives. And that's really just that-that difference between Q38 and Q20. 

So depending on the context of what you're asking in assays that are employing these sort of techniques, you can use both simplex and duplex. Duplex you may want to employ where you're interested in what exact bases are converted or what bases aren't converted. But if you're just doing region-level searching or mapping, simplex is probably sufficient, and that will allow you to start to leverage the large-scale data that comes out of a 4-hour run. And so when we talk about flexibility, it's the library prep combined with the sequencer really for the application, right? You need to tailor what mode of sequencing we're going to be using depending on what application you want to do. It's the right technology for the right choice. 

And so we'shown compatibility with both bisulfite and EM-seq, and really going forward, we're going to generate native library preps that work with this. So working with those on-market methylations and not go through this conversion step. The conversion step is just really a nice start for libraries that exist, but long-term, we want to get as close to the sample as possible, and that's by using the AXELIOS adapter systems. We also will expand a little bit more on the EM-seq collaboration we're-we're discussing to really talk about the biological implications and what AXELIOS brings to that sort of analysis. 

The next vignette I want to talk about is library sample type mixing. Sean and Jorge showed SBX duplex alone. In previous presentations at AGBT, ASHG, we've seen simplex alone, we've seen single-cell alone, we've seen bulk RNA alone. But we realized not every lab is going to have enough of one workflow to fill up their single 4-hour run or whatever time they choose to use. And so we want to show what's possible with the system. And so in this initial experiment, we worked with Dr. Graham Wiley at the Oklahoma Medical Research Foundation, where he gets samples of all types. He gets model systems in birds, model systems in zebrafish. He does a lot of human as well, and he does both whole genome, so DNA-based, as well as RNA-based. And he said, "Well, I'd really like to understand what are the biases I see on the AXELIOS system versus what I'm used to doing in an SBS-basis system." 

And so here he combined eight different libraries together. We have bulk RNA, we have three-prime 10x, and so that'll have the poly-A tail, which in a SBS sequencing world, you really need to be careful how much of that poly-A tail you put in there because it starts to drop your quality scores during sequencing. 

You have to add things like PhiX into it, which start to also take reads away from what you're looking at. And what you can see here is that we had a pretty wide distribution of read length, ranging from about-well, we'll see on the next slide, it's like 350 up to about here on the tail, 700. 

This again was using our prototype conversion workflow because we wanted to compare directly to the pool that Graham had put on to the 25B flow cell. And so what we did is we took that pool, which consisted of all eight of those libraries, and put it through the conversion method, and then left the SBX handle present so it could interact with our system. And what you can see, we did two separate runs just to start to look at repeatability in this system. On the first run, we got about 17 billion, 18 billion raw reads, and on the second run, slightly less, around 15.9 billion. You can see we have a couple steps there. We have very strict quality control right now on the 10x ones where we're really only counting full-length reads of the conversion. And so to get to a final read, you start to go through some steps of churning off reads that may not go the long-the full length. But those are all useful, and as the data's dug into more, we'll begin to understand where those reads can be used and where they can't. This was a-a 5-hour sequencing run, 1 hour for setup and 4 hours for sequencing. And then the match data was generated on a NovaSeq X 25B 300-cycle kit. And then the analysis was done by Graham. 

So here, I don't want to focus on too much of the details here because of another busy slide. So if you take pictures, you can check. But what you can see is that there's a distribution in the median read length from about 310 bases to 460 bases of the library. During the Xpandomer synthesis, we have the ability to tailor that protocol for different lengths, and in this experiment, we tailored for the longer read lengths because there was a question on the RNA samples of what happens when we're getting the longer data out. So we really wanted to tailor towards that.

And what that meant is that when we look at the percentages of the libraries recovered on the AXELIOS runs, you're going to see a bias for those longer read lengths and you're going to see less of the shorter read lengths. And you can really see that in the 10x_A and B where we're recovering 15% when the pool only had 8%. And then on the mRNA_2 where we were targeting 35% in the pool, but we only recovered 24%. This is really just to stress that, due to the synthesis where we're trying to get longer reads out. 

And so now I just want to briefly take you through some of the highlights from some of the sample types just to show you how performance looked. You can see here, looking at the mRNA samples, we detected very similar numbers of genes both in both AXELIOS runs, as well as compared to the original pool. We had very high mapping rates across the board, as expected. And then the rest of the metrics also were equivalent. If we look at the 10x samples, all of them are behaving as expected. We're not going to go through all the numbers, if anything, where we're recovering a bit more cells in the AXELIOS runs, but this is really due to generating a bit more data. These weren't downsampled to the exact amount. There were about 80 million in each.

If we look at mapping, we're seeing equivalents across the board. This is, you know, both intergenic as well as intronic, exonic, as well as total mapping. And so what I want to take away from those is that you can start to combine libraries of different types together. We're working with other collaborators in this space to start to stress the system. This was about 150 base pair difference. We want to show what's possible both on the shorter end, as well as the longer end, and then see where's the breaking point for what you can do.

But what the flexibility of the system allows you in your lab to do is say, "Hey, I don't have to hold these samples for multiple days. I have five of one type and two of the others. It's okay to start to pool those together." The real learning, though, is that the current pooling rules that you're using are built for specific sequencing types, right? You're going to have to develop new rules, and maybe if you're biasing your length towards the longer side, you may have to dope in a little bit more of the shorter ones to recover what you want. But this is the same stuff that goes on in the SBS world today. But SBX is fundamentally different, and going through the synthesis process, we have to tailor some of the protocols we have there.

And so over the past year and a half, Mark, Mitu, and the team has shown a lot of different applications that we're working on. We highlighted a lot of whole genome sequencing today. We have bulk RNA, we have work done with short RNA. Uh, Aziz at the Broad has highlighted a lot of work in the single-cell space. We have spatial ongoing.

But as Mitu said, we're an open ecosystem, and we're going to continue to develop these applications as we go along both internally with application notes and white papers, but also working with the other players in the field to make their protocols compatible with SBX. And so, Proteomics is next. We've shown Olink, we-we talked about the collaboration, but we're also working with a company called Pixelgen, which is looking at spatial proteomics, and Alamar, which is another PTD company. And so what you can do is apply AXELIOS where you're doing it, but you're going to have to tailor it for what it is. But we're really excited for what's to come. Thank you.