Welcome, and thank you for tuning in for a streamlined workflow for CNV analysis on whole exome sequencing. My name is David Schlesinger, product manager at Sophia Genetics. I'm here today to introduce our two speakers. First is Dr. Clay Morrison from Twist Bioscience. Clay received his PhD in developmental biology in 2008 from Baylor College of Medicine in Houston. He began working in next-gen sequencing fields shortly thereafter as a product project manager for a contract research organization. Clay has been with Twist since 2018 as a field application specialist focused on target enrichment. It's also a pleasure to welcome my colleague, Dr. Emily Paul. Emily is a molecular biologist with a specialty in genetics. She earned her PhD at the State University of New York at Albany and was subsequently a fellow at the Wadsworth Center at the New York State Department of Health, the Boston Children's Hospital, Harvard Medical School, and the Dana-Farber Cancer Institute. She joined Sophia Genetics in 2017 as an expert and is now the director of our fabulous subject matter expert team here in North America. This slide is just their disclaimers and disclosures in that Sophia products are for research use only. And as many of you may know, here at Sophia Genetics, we provide tech solutions for healthcare institutions in the areas of genomics, radiomics, and clinical trials. And now I'll turn it over to Clay. Thank you very much. My name is Clayton Morrison. I am a field application scientist with Twist Bioscience. And it's my pleasure today to talk to you about how the uniformity of our capture technology and the performance of our capture technology fit very well with Sophia Genetics' analysis platform and particularly their ability to call CNVs. What does Twist do? At its heart, Twist is a DNA synthesis company. But over the time since we've been founded in 2013, we've branched into several different areas. We can make clonal genes for customers, oligo pools. We do combinatorial libraries with very specific ratios of the different variants. We do NGS, which is what I will focus mainly on today. We also have done some groundbreaking work in DNA data storage, as well as in the biopharma space. So what sets us apart is really at the heart of our manufacturing process. We generate our oligonucleotides on a silicon array, which allows us to make close to a million oligos on a single array. We separate these oligos into groups of 121 devices. We call these in an individual cluster. This cluster has about 9600 clusters on this plate, and each one has 121 individual oligos in it. As we're doing our manufacturing, we can QC during the manufacturing process to make sure each of these oligonucleotides are being built, and if there are any issues with any of these clusters during generation, we X out that cluster, regenerate it, and add those into your pool so that you never have any issues with probes dropping out. Some of the primary benefits of the Twist target enrichment system. The first is exceptional performance. We use double-stranded DNA probes that have a high capture efficiency in order to capture as many of the unique molecules as possible in your library of interest. This leads to a reduction in the amount of bottlenecking that occurs in a capture process, leading to more uniform data and better ability to call variants. We also have a great amount of flexibility and are able to set up projects and panels in whatever way customers are interested. We also have worked well to maximize the sequencing efficiency, again, through our uniformity. So when doing target enrichment, what we are trying to do is get even coverage over the target regions of interest, and we want to minimize the times when we over-sequence something, when we under-sequence something, or when we pull down regions that are not what we're looking for. All of these lead to wasted sequencing utilization. But how do we achieve this? All oligonucleotide providers use the same basic underlying chemistry in order to generate their oligos. This type of chemistry is very uniform across the GC spectrum. So low GC to high GC, we get excellent representation of the different nucleotides. The difficulty comes in for target enrichment when you have to amplify that material. If you perform standard amplification, even on a very uniform synthesis, you will see GC biases. In the sweet spot of the GC content, you're going to get a lot of coverage, and at the tails, you see a lot of reduction of coverage. With TWIST, with our proprietary manufacturing process, we're able to undergo proprietary amplification that gives us a very uniform probe pool across the GC spectrum. By eliminating the bias that occurs during this amplification, we can really help our ability to cover those regions of interest. In the final data, you can see that the sequencing reads covering target regions of interest compared to a competitor are indeed much more even. So with TWIST, you see that low to high GC range, we are getting similar levels of coverage, whereas with the competitor, the low GC regions are having trouble and the high GC regions are overamplifying. We like to look at our captures with a metric called the Fold80 base penalty. This is a uniformity metric, and the point of the Fold80 is to measure the amount of over-sequencing required to achieve a given level of coverage. For Fold80, that means how much additional sequencing would I have to do in order to get 80% of my bases to the current mean level. If we take a look at these competitors, a 1.9 and a 1.7 for the A competitor, this means this sample would need 93% more sequencing. This sample needs 71% more sequencing. The I competitor also needs about 70% more sequencing to bring the distribution up to the average level, whereas TWIST is down at 1.34 with their Fold80 value. Interestingly, this is right about the same level that you would get if you were to do whole genome sequencing and then analyze the uniformity of your target regions of interest. So essentially, using the TWIST capture technology, you don't introduce any bias into the capture system. Another metric that we find very important is the duplicate rate, which is also tied into the HS library size. The duplicate rate, as many of you know from your next-gen sequencing experiments, just occurs when reads have the same start and stop point. So at TWIST, since we're able to pull down so many molecules during the capture process, we recover far fewer duplicates when we get to the end of the analysis. With all of these duplicates removed, that just means that far more of this data is going towards what you actually want and not being thrown away. If we take a look at the HS library size, we see a similar effect, which is that despite the fact that these competitors went into their library preps and their captures with more molecules, TWIST generated over six times as many unique molecules at the end of the capture process. So we really are able to pull out that complexity out of that capture system. The benefits of having this great uniformity and then the excellent duplicate rate is that far more of our data ends up being on target and what you actually want. So if you compare the competitors at 20 and 30x levels of coverage, you can see that pushing the coverage levels up starts to show drop-offs in regions of interest. Indeed, with the A competitors, if you're interested in a 30x coverage level, you've lost almost a third of your bases, so you really can't call these variants very well. As you move to the I competitor, it's doing a little bit better, but still at the 30x coverage level, you can see that it's down below 90%, whereas with TWIST, we are able to maintain high levels of 30x coverage even when we 8-plex the samples. The benefit of this really is that you have more of your bases covered at a genotypable depth. You don't have to do additional sequencing to bring up bases and you're able to get more bang for your sequencing dollar. At TWIST, we have two different versions of the exome. The streamlined core exome was our first offering and it was designed off of the CCDS database. CCDS is a conservative database focusing on the coding sequences, and we cover well over 99% of that in our core exome. Our comprehensive exome adds content from the RefSeq database as well as GenCode. This is about a 10% increase in the size of the panel, but brings up coverage of these two important databases above 99%. Depending on whichever type of exome you want, TWIST can help you get there. We also have the ability to add spike-in content to either of these exomes to make it look exactly the way you want. Indeed, this is something we do with customers on a regular basis. We pick one of these, either the core or the comprehensive exome, and then add whatever content you're interested in. A benefit of this is that we can add the content to be even with the coverage levels that you're achieving in your exome, or we can make it so that your spike-in regions achieve higher or lower coverage depending on your needs. Sometimes people want their spike-ins to have, say, twice as many reads as the exome. This is something that can be easily tuned using the TWIST technology. If we zoom in and take a look at some of the regions that we cover with a spike-in, you can see that the addition of spike-in probes, which is down below, you're still able to get excellent coverage over those newly added regions in very similar coverages to what was present in the initial panel. Our spike-in approach works very well to round out your panel into whatever you're actually interested in looking at. At TWIST, we have reagents across the spectrum from library preparation all the way up to sequencing-ready libraries. We can accommodate enzymatic fragmentation, mechanical fragmentation. We have a few different types of adapters. On the enrichment side, where we use the oligonucleotides to pull down the regions of interest, we have a standard overnight hybridization approach, and we also have released a FASTHIB approach, which can be completed in the hybridization step in as little as 15 minutes. We have our exome catalog panels and a variety of custom panels. And recently, we have also released a methylation analysis system as well. So that is another area where TWIST would be able to help you. So if we come back to what is it that really kind of sets TWIST apart, it's the exceptional performance of our system. The double-stranded DNA probes with their high capture efficiency are able to pull down at the base twice as many unique molecules as a competitor that's only capturing with single-stranded oligos. Our flexibility, we are able to customize and generate whatever types of panels you're interested in running, and we can turn these around quickly. Generally, we can get you a new panel in three to four weeks. Maximize sequencing efficiency, if you get more coverage over your target bases of interest with less sequencing, which is what I showed, you are then able to maximize the number of sequencing samples going on to your sequencer. Together, these different facets of the twist performance are really able to maximize the uniformity of coverage over your targets and get you the best bang for your buck on your sequencing. And with that, I would like to then pass off to talk to Sophia Genetics, Dr. Emily Paul, to explain to us about how the SOFIA system and, in particular, the CNV analysis suite can be a powerful tool when combined with TWIS uniform capture technology. Thank you very much. Thanks, Clay, for that introduction and for talking about that importance of coverage uniformity in a great assay. And actually, that will be really relevant for us talking about copy number variation analysis on whole exome sequencing, specifically using the TWIS exome dataset. So at SOFIA Genetics, David touched on this earlier, we are a health tech company providing solutions for health care institutions. And we have experience looking at multiple combinations of assays, sequencers, applications, sample types, all of these things that have different noise signatures. And through our universal platform, SOFIA, we're able to see what these signatures are and better understand the signal amongst the noise. Now, whole exome sequencing is particularly challenging, as we all know. There's a lot of benefits to it. But with more data comes more challenges. And these include achieving sensitive CNV detection. And then how do we filter through this data set? As a positive, all of that data is exactly what you're looking for. It's a motivator for doing these whole exome solutions. We want to be able to see all of these different novel disease-related genes and have a streamlined lab workflow. So definitely a reason to look into this. But how do we do it effectively? Well, we can look at the 21,000 genes that are uncovered by the TWIST human comprehensive exome by TWIST and able to look at the data we generate from there. Now, I'd like to take you through three parts of the process of looking at the data. Clay took you through generating that TWIST exome data set. But then I want to take you through data upload and management, secondary analysis, and followed by tertiary analysis through the SOFIA platform. Now, we want to be able to analyze our data at any time. We know that it's important for labs to have access to their data and what they need to see. So we envision this as a cycle where we're enabling labs to take control of their data. So the health care institution is able to directly generate their raw genomic data, enter that into SOFIA DDM, and use the platform to generate an interpretation or to use it to follow through on the secondary and tertiary analysis to a report. Now, as I mentioned, we want to help the labs take ownership of their data. And that's exactly what we do. Data and security are inherent within what SOFIA has as a platform. We work to make sure that they have access to and can analyze their data at any time with the confidence of our ISO 27001 certificate. Now, the second part is secondary analysis. So how do we identify those copy number variations within the exome data set? And this is actually an incredibly important part of what a lab may do, particularly those labs that are going to look for disease-causing variants within copy number variations. In fact, 5% to 10% of those disease-causing variants are C and V. So you really shouldn't be missing those. And then we have that high resolution. Because by having copy number variations as a portion of what you're able to do, you can increase your diagnostic yield by up to 35%. So really important to be able to see these in a challenging situation. Now, one of those challenges, one of the primary challenges is understanding signal-to-noise. And in fact, signal-to-noise in a very large data set. The larger the data set, the more noise. So how do we do that? Well, we make a couple of assumptions. And the first assumption is that no C and V is shared by a significant number of samples in the same run. Often C and Vs are something that we see as unique or less reproducible than a SNIV, for example. So we'll look at the reference genome using exon A and exon B here on the left of the screen to be able to see what the assay looks like. We have sample one and a separate sample as well as a regulatory control. We can see the lack of coverage suggests that there is a deletion in exon B. And what would that deletion look like within the SOFIA platform? In fact, we can see that here on the right-hand side of the screen where the three dots towards the bottom show a loss in one copy variation. A loss in one copy number. In fact, the gray dots also represented in this plot show the noise that is present in any normal assay. And we see the distribution of that noise. So how do we look at the noise itself? Well, we have to know when to reject the noise. In fact, we can see here on the bottom portion of the screen what a noisy sample looks like with all those Xs and with all the discrepancies between the background itself. And noise comes in at multiple steps of an assay. It's not just the extraction. It's not just the library prep, capture itself, sequencing. All of these are contributing factors to understanding that noise. So we must take them into account when developing our C and B algorithm. Because without that, we will never be able to see a single copy deletion over three exons in an exome-based dataset. We must understand how to tease that information out. In fact, to highlight some of the noise issues that we can see, if we have greater than 15% noise in a single capture pool, we're unable to call the copy number variations. Now, there are different ways that we filter this out, but there are dual normalizations to both the samples and to the locus itself. This also allows us to call long-range events. Long-range events are incredibly important to be able to detect as well as the smaller ones. And here we can see in the upper portion of the screen, the wild type representation in an IGB screenshot looking at approximately 54 exons. Now, we see those 54 exons in the mutant sample just below it, and we see a decrease in coverage. We go from about 1200x to 750x. And we also see that representation in the plot below, where we see the deletion in two full genes, but it doesn't touch the gene next to it, which is the IGB screenshots we have on the right. So we know that that loss of coverage is actually important. The longer-range events, in fact, we can have higher confidence in those because they span multiple probe touch points. And this way, we're able to confidently report the loss of the exon. Now, that's not to say we, of course, can't include the shorter-range events. So the longer-range events, as I mentioned, are covered by more probes. But the shorter-range events have a higher chance to be covered by the individual probes as well as the individual reads here. So we look at the reads that we're going to see, and we know we can actually visualize the information we're able to detect. So the combination of both the short- and long-range coverage really helps us have confidence in a low number of exons as well as broad-reaching copy number variations. And so to also state this covers SNBs and indels. Indels are also a change sometimes in the copies that are present. If we have a deletion, for example, we want to be able to discern those. And we're going to be able to do this all within one type of sequencing assay. Now, how do we have confidence in this? Well, in fact, we were able to look at 187,000 region exons for copy number variation within these assays. And we wanted to be able to benchmark this solution. We wanted to be able to be confident in what we were doing. Our original development came from a hereditary cancer solution where we were able to use 139 MLPA-confirmed CNV samples. And then we extrapolated that and then tested it further on samples that had been processed with the TWIST comprehensive exome. So we have another 25 samples to be able to understand the scale of the coverage that we need to be able to gain resolution on single exon or multiple exon deletions and duplications. And you can see that presented on the right-hand side of the screen. So we can see with 160 million reads, we're able to cover quite a bit, as you might imagine. But even as we get into 80 million, we're able to detect the single and dual exon deletion and duplication events and able to go much further. In fact, we get in the 2 to 5 range very fast, even with 40 million. So we do start to gain confidence as we increase the read number. And to come to that coverage uniformity, this is a crucial part of how we're able to detect this. We want to be confident in what we're seeing in every run. And by having a assay that allows for minimized noise for a consistency throughout, we can then confidently call those copy number variations. And just to show you the sensitivity that we're able to achieve, in fact, we get greater than 90 percent sensitivity, as you can see in the plot on the left, where two consecutive regions or exons can be detected. So anything with some sort of deletion or duplication can be done with 80 million reads. So really a nice resolution on this assay. Now, just a quick note about tertiary analysis, because we have to be able to see these variants. It's great to have what you have, but how do you know what is there? And we mentioned that as a challenge at the outset. Whole exome sequencing can be a difficult thing to do. Well, we want to provide the tools within the SOFIA platform to be able to streamline that variant detection. Now, thinking of our three categories that we mentioned, SNIVs, indels, and copy number variations, we want to be able to filter them from the original, let's say, 25,000 variants to anything as likely or potentially pathogenic through to relevant disease genes. Finally, to have different cascading filters, which is a function that allows you to put scores of population frequencies, something from the generically available databases like ClinVar, et cetera. And then finally, possibly a familial variant analysis, depending upon how many sequences are available to support that individual analysis. So a lot of tools to make the burden of large-scale datasets easier. And a report is, of course, available to be able to provide this result to others. And we have confidence in the data that we are collecting. We have an increasingly growing global community where we have seen 650,000 genomic profiles, and we're averaging about 18 new genomic profiles a month, which equates to about one analysis every four minutes. So the more we see, the more exposure, the better we can fine-tune our algorithms to support the data. And with that, I'd like to conclude that we have analyses possible at any time. The advanced analytical performance to identify the multiple types of variants, specifically discussing the sensitivity of copy number variations, even down to a single exon level resolution, and the platform to help make sense of all the data that you see. With that, I'd like to thank you for listening to our session, and I'd like to thank our speaker, Clay Morrison, from TWIST.

Twist Bioscience

CNV analysis

whole exome sequencing

target enrichment system

SOFIA

collaborative potential