Earnings Call - QIAGEN NV

Hey everybody, I'm Ben Busby. I've been a computational biologist for 20 years. Now I'm the omics alliances manager for NVIDIA and I really look forward to expanding this collaboration with you. So at NVIDA we have an amazing solution that puts GNNs in front of LLMs and co-catenate some beddings for graph rag and really what we were looking for was a partner with a large very precise well curated knowledge graph and QIAGEN was an obvious initial partner for that and working with QIAGEEN has been really amazing and easy to do in terms of both pushing the science and marketing with those colleagues. Hi there, I'm Venkatesh Muktali. I'm a director of product management at Tyogen Digital Insights. I manage our knowledge graph product portfolio. We are at BioIT today, 2026 in Boston, Massachusetts and I'm very happy to talk about something we're launching at this event. It's called the Tyogen Discovery Platform. So Tyogen discovery platform is a full-stack product which in the sense that it goes straight from a foundational layer of data which is knowledge graphs that we've worked on over the past 20 years curated and been cited more than 70 000 times in peer-reviewed literature so that becomes our base and then on top of that we are building mcp servers and additional services including apis that then provide all of this data to our customers agents as well as our own applications that sit at the top layer and so entirely in entirety these three-way architectures entirely decomposable where you can bring in your own components your own ages and take advantage of the data that we curated over the last 20 years so we are collaborating with nvidia on very specifically a problem that a lot of the community is trying to answer or solve which is given a complex question in biology do you come up with contextual and precise answers and that is an unsolved problem for most part today and so uh this collaboration attempts to solve that problem by bringing together a conference on bionumo platform and then combining that with our industry leading manually curated knowledge graph to then come up with a specialized retrieval augmented generation architecture that provides precise answers And so what we've been able to do is we've been able show benchmarking results that provide 5 to 10x better performance as compared to standard retrieval augmented generation pipelines. And so this is just a starting point for the collaboration. The ultimate goal is to then provide to our community full architecture that combines NVIDIA's accelerated computing with our industry-leading data and as well as algorithms to come up with contextual answers that our pharma collaborators can rely on, can defend, and can then ultimately move the drug discovery pipeline forward. So BioNemo is really a collection of models that NVIDIA has accelerated and put out there and uh really we're very interested in uh talking to uh kyogen about their amazing uh knowledge graph capabilities because honestly we uh we believe that knowledge graphs can help uh validate uh many of the models that folks are running uh as well as um as well as to help make better prompts, do literature mining, and make answers more accurate. Chat agents have become very popular. Everyone loves the conversational way of interacting with large language models and getting answers, and this has been proven to across the entire BioPharma value chain. Whether you're a biologist, whether you're data scientist, you're C-suite, everyone wants to ask questions from different points of view but often what's happening is people are concerned about hallucinations people also equally concerned and this is perhaps something that doesn't get as highlighted is that you can ask a question in oncology and sometimes you may get an answer that is not relevant to oncology at all which is lacking that contextual element in your qna systems is something that is an unsolved problem one of the ways that teams have tried to address this is through retrievable augmented generation or RAG. But standard RAG still only gives you answers from underlying data. And if the data is sparse, you don't get any answers. In addition, oftentimes, the elements don't give the full context by just throwing different parts of the graph. And so that's what we are addressing in this NVIDIA collaboration, where we are we are implementing a special form of RAC called the G-Retriever, which is part of the BioNemo offering, or BioNEMO platform. And in this architecture, we provide specific parts of the graph that are relevant to your question in a Q&A system, and that improves the answer, so it's more context. And that ultimately gives you more confidence in your Q&E systems that your organizations will use. So I think knowledge graphs are pretty much the obvious solution in terms of validation of models and really in terms validation of unstructured data and their contribution to models. One of the things that you can do with the gRetriever system is to actually build a new knowledge graph with previously completely traversed subgraphs from an existing knowledge graph and then adding unstructured information. That's really one of the things we're very excited about. Another thing we're excited about in this sort of economy of model licensing, the idea of being able to track for any given prompt and response where that data came from and which data is answering which questions. I think that's really going to contribute to the entire ecosystem. So the PyTorch Geometric GitHub repo is actually free and open source, but to get access to to QIAGEN's knowledge graph. You're gonna wanna talk to QIAGEN and what they've been doing is enabling a really easy to use version of the PyTorch Geometric GraphRag right in front of that knowledge graph so that it's really available to use out of the box with the best in class knowledge graph already there. This ultimately builds into the discovery platform that we are launching at BioIT. And this platform, along with the GVTable components, will truly democratize access to factual data and textual data across the organization and help us build better agents that can provide better ground truth and answers that can then push the space forward. And that ultimately gives you more confidence in your Q&A systems that your organizations will use. So I'm very excited about the next steps here. So we are only starting by showing the benchmarks and showing how good the results could be. Our next step involves identifying and working with several pharma partners to implement this G-Retriever framework across different data payloads and different functions. As a next step, what we would do is we'd offer it an mcp server that any anyone can use as part of their workflows and start getting uh factual contextual answers uh this ultimately builds into the discovery platform that we are launching at bioit and this platform along with the gpt components will truly democratize access to factual data contextual data across the organization and help us build better agents that can provide better ground truth and answers that can then push this piece forward. But to get access to QIAGEN's knowledge graph, you're going to want to talk to QIAGEN and what they've been doing is enabling a really easy to use version of the PyTorch geometric graph rag right in front of that knowledge graph, so that it's really available to use out of the box with a best-in-class knowledge graph already there. So in terms of what we're doing next, I'm really excited about being able to track where data comes from, and then also thinking about explicit validation of multi-hop queries and multi-hot models. We're able to generate prompts. That's extremely helpful. And then again, being able to generate new knowledge graphs and building validation systems and also comparative systems between models with knowledge graph integration is really the next frontier. and I'm very excited about it.