Earlier this month, I traveled across North America and the UK to meet our 340 team members. Every year in January, we kick off the year with our “Roadshow”, where I meet our team in person and share our accomplishments and learning from the previous year and where we are going next. This year, you could feel even more excitement in the room than in previous years as we are accelerating our efforts to integrate even more Generative AI (GenAI) capabilities into our ASCEND Platform.
We believe the greatest opportunity of AI and GenAI in drug discovery is to unravel the complexity of disease biology. We are excited to increase our efforts as we work towards our mission to improve the speed and quality of R&D, helping bring new medicines to patients faster. In this blog post, I will share our perspective on GenAI and how we plan to augment drug discovery workflows at scale.
When it comes to pre-clinical research and development (R&D), there are two main ways to apply GenAI: drug design and disease biology.
Until now, most AI R&D funds have been directed toward drug design, yielding remarkable advancements. We’ve witnessed notable progress in precision targeting and predictive modeling, enabling scientists to hone in on disease-related biomolecules with unprecedented accuracy. More recently, the integration of GenAI has enabled scientists to efficiently generate novel molecular structures. AlphaFold’s groundbreaking protein folding prediction capabilities exemplify the potential synergy between GenAI and drug design. Another example is Sanofi’s recent investment in BioMap's protein mapping technology, which has the potential to accelerate Sanofi’s drug discovery by precisely targeting proteins associated with specific diseases, streamlining the identification of promising therapeutic candidates.
With these significant advancements, it’s becoming increasingly clear that GenAI in the drug design market is swiftly approaching maturity. The biggest reason for it maturing rapidly is because drug design has predefined rules. Disease biology, on the other hand, does not.
Today, disease biology is more complex than ever. And getting the biology wrong is the number one reason clinical trials fail. That’s why applying GenAI to unravel the intricacies of disease biology is not only the biggest opportunity in AI today, it’s also the unmet need in transforming drug discovery.
But the question remains: how do we do it?
There are a few players in the market right now working on solving this problem. Some focus on combining distinct datasets together, others concentrate solely on preclinical or clinical data, and a few are establishing comprehensive laboratories for wet experimental work. Many of these companies aim to transition into AI-driven biotechs, forming collaborative yet competitive relationships with pharmaceutical companies—a dynamic often referred to as "frienemies". We believe that while this is extremely valuable for the industry, it will not solve the problem at scale.
BenchSci, on the other hand, is paving a different path forward. Over the last eight years, we have been teaching a computer to understand biomedical research the way a scientist would, but with AI. Ultimately, building an AI Assistant to augment the capabilities of preclinical scientists in comprehending the complexities of disease biology. What sets us apart from biotechs is a strategic decision—choosing not to leverage this technology for in-house drug development. Instead, our focus centers on supporting pharma programs as they discover drugs. Aligned with our mission to accelerate R&D at scale, our AI Assistant provides robust support to every drug discovery program across various therapeutic areas, ushering in a new era of efficiency and innovation in the pharmaceutical landscape.
ASCEND is a science-first and tech-deep disease biology GenAI platform that acts as an assistant for scientists and decision-makers in pharmaceutical preclinical organizations.
We understand that despite scientists completing millions of experiments to understand the different building blocks that make up a disease, it is not humanly possible to connect them all together, which is why disease biology is extremely intricate and complex.
We developed our proprietary multi-modality Machine Learning (ML) models, including specialized vision and NLP algorithms, Large Language Models (LLMs), and knowledge graphs for biological ontologies. Altogether, they build a deep, comprehensive, and evidence-based biological understanding of diseases.
Our biology-specific multimodal AI means that our ML models focus on understanding experiments from both text and figures. This results in powerful Generative AI without hallucinations and with scientific explainability.
Using AI, and our proprietary Machine Learning (ML), to read and scan figures, charts, and graphs, scientists can uncover data and evidence never before correlated or discovered. Using curated ontology datasets, ASCEND makes connections across experiment outcomes to create the first commercially available, unbiased, and evidenced-based map of the underlying biology of disease. We also understand the value of providing a holistic view of disease biology, which not only includes the world’s scientific findings but also internal findings from our customers. For this reason, we developed a technology to enable the secure integrations of our pharma’s unstructured proprietary data. By combining the two data sets, external and internal, we created the most robust data foundation in the world, enhancing discoverability and innovation. It helps scientists end-to-end throughout the preclinical process to discover biological connections, dramatically reduce trial-and-error experimentation, and uncover risks early to move the most promising projects forward faster.
ASCEND increases the productivity of the pharma drug discovery pipeline by supporting three main workflows: target and drug due diligence, experimental design and validation, and translational workflow. We achieve this by identifying the best target, coming up with the best idea, and testing and validating it quickly, ultimately increasing the chances of its translation.
One example of this is our work with Novartis, specifically relating to our work together around their experimental design workflow. In their 2021 annual report, Novartis stated they saved $14 million USD, emphasizing that “BenchSci's technology has helped Novartis scientists select the best antibodies and reagents, cutting down on expensive and unproductive experiments and accelerating projects by months.”
While we’ve been using Large Language Models (LLMs) for years, we are enhancing ASCENDs’ capabilities by accelerating our investment in GenAI. We’re doing this across two main areas. The first is data foundation—to improve our depth and breadth of scientific insights and evidence even more. The second area is continuing to evolve the user experience by enabling summarization, conversational AI, and explainability. Our platform is designed by scientists for scientists. As ASCEND continues to evolve, we are meeting scientists where they are. Simply put, we are building our platform to mimic how a scientist would understand and extract biomedical data. This user experience empowers scientists to effectively pick the right experiment, find the best path that can impact a disease and set that up to be more successful for clinical trials.
We’re fortunate to work with 16 of the top 20 pharmaceutical companies, helping their teams of scientists as they work towards breakthrough scientific discoveries. This collaboration requires a high-fidelity solution, which is why we adhere to a rigorous validation process as a pivotal step before integrating any technology into our platform.
One example of this is our recent integration of Google MedLM’s technology. Late last year, we closely collaborated with Google’s Product Development team and assessed, validated, and provided valuable feedback on the model as it moved to general availability. This collaboration allowed us to further strengthen our partnership with Google (whose AI-focused venture fund, Gradient Ventures, is also an early investor in BenchSci) and accelerate our exploration with market-leading LLM solutions while staying loyal to our best-of-breed, fit-for-purpose LLM principle that is core to our GenAI tech strategy. This approach allows us to develop a growing arsenal of multiple models, each used where they shine the most, all in the service of combining it with our biomedical domain expertise and bringing innovative solutions to create further excitement in the pharma market and deliver unique value to our customers.
Equally crucial in our validation process is the feedback we receive from our Scientific Advisory Board (SAB), which includes experts in life science research and development and machine learning technology, and our customers. As we evolve our platform, their lived experiences ensure we can continue to build for the needs of pharma organizations, addressing the pain points for scientists and transforming drug discovery.
Recently, Jensen Huang, CEO of NVIDIA, said "There is no more meaningful and exciting place to be than the intersection of biology and AI,” and I couldn’t agree more.
Drug discovery is wide and deep and as a result a complex and challenging field that requires a substantial amount of time and effort, but at BenchSci, we believe we are working on solving the biggest problem in one of the most impactful industries in the world.
Understandably, there’s a lot of hype and excitement around AI and GenAI. However, our focus remains steadfast on the real impact and tangible value these advancements can bring to the world of preclinical drug discovery. This involves not only developing robust building blocks but also continuing to be a leader in innovation, expanding our ASCEND platform with even more GenAI features such as conversational AI and explainability.
While the field of biology and AI is still in its early stages, we believe, like how every car is going electric, that every pharma organization will harness the power of AI and GenAI in their R&D efforts, which has the potential to revolutionize our understanding of disease biology and our ability to treat diseases.