Bob Chen: Hey, everyone. I’m excited to be here to talk about how we’re enhancing discovery with OmniDeep. In May, we launched OmniDeep to highlight for our partners, the suite of in silico tools that we have for therapeutic discovery and optimization. These tools are woven through our technology stack to help us create diverse repertoires, to help us screen millions of cells. And they help us deliver the right antibody. Our goal is to use these in silico tools to streamline and assist drug discovery. We want to make our capabilities ever more efficient and effective for our partners. OmniDeep is built on four fundamental pillars. It’s based on deep repertoires from our animals. It’s based on deep screening with proprietary hardware and software.
It is based on deep sequencing, which collects in-depth information and is based on deep learning tools that help us understand this information and make insightful decisions. I want to emphasize that the input data of OmniDeep are the deep repertoires that we get from our animals. These repertoires are generated by biological intelligence. This is the interplay between rational genetic design and powerful in vivo process. For example, on this image here, is a illustration of what repertoire space may look like in three different animals. They are immunized with different variants of the same targets, two protein variants and one genetic immunization with either mRNA or DNA. And you can see the repertoires look different between these animal.
And so biological intelligence allows us to create vast and diverse antibody repertoires within an animal and across animals. This is a large space, and we need to bring unique tools to be able to understand and tap into what we create. And here we can bring the xPloration platform. This is our AI-driven deep functional screening platform, and this platform is built around this chip shown here. So this chip, which is about the size of my palm, has 1.5 million 40 micron features in it. This is a unique through-hole format. And in each one of these microcapillaries, we can form an assay of some sort. Very often, we perform binding interactions with florescent antibodies. And then we bring in a suite of various machine learning hit detection models that allow us to automatically look at every single one of these cells and label the target of the hits of interest.
Once we have a list of cells we want to recover, we use a precise and proprietary laser recovery method to recover the cells very quickly, which are then fed to a single cell bar coding workflow or a full sequencing workplace. So what we’re doing is integrating biological intelligence with AI. We feed biological intelligence into these deep screening and sequencing tools, which enable us to have large-scale data collection. All this information is dumped into proprietary databases, which then feeds into deep learning models, which we stack on top structure-based design tools to really have high-quality sequences. These tools and this ecosystem allow us to better mine the diverse repertoires that we have. So biological intelligence and exploration creates a synergy, which generates a large amount of data.
And OmniDeep is the key to navigate this data and to empower what we call large-scale antibody discovery. To illustrate, what I mean is we’re going to talk to our case study. In this case study, we immunized three OmniFlic animals and we screened over 27 million cells that led to over 3,000 positive binding events and over 1,300 unique sequences that span over 124 lineages. We like to visualize repertoire space with the spot here. In this spot, this is the top 30 lineages across the screen I just mentioned. And you can see that every one of these spots represent unique antibody sequence. So this illustrates the scale at which we deliver to our partners. Now to go from a large pool of sequences to therapeutic candidates, we need to have a strategy, and we need to make some decisions along the way.
Traditionally, you may look at perhaps the lineages that are boxed in the squares here. They’re boxed because they are likely to be both protein and cell binding but now with OmniDeep, we can apply some in silico tools to intelligently prioritize and select candidates from this large pool. So how does that work? Well first, we can take the high quality input data, the xPloration hits and perhaps assay data, affinity value from select clones and we can put in all of the NGS data from the immunized animals. So we can put in real sequences that exist in animals and also very high confidence data on a few select clones. And we can feed that into deep learning. Very often, we use a type of model called a variational autopilot. And the goal here is to extend the insight on those confirmed hits to be able to infer the function of untested clones.
