Over the past few decades, we’ve witnessed the power of machine learning (ML) and artificial intelligence (AI) in applications such as language translation, stock trading and space exploration. While many of the more public-facing uses of AI have been driven by the tech sector, scientists around the world have also been working to harness it to ask bigger questions and address previously intractable challenges in human biology and disease.
Advancing AI tools, such as ML, in drug discovery and healthcare is more important than ever as drug developers are moving beyond the universe of familiar targets and are tackling increasingly challenging ones to treat more complex diseases with high unmet need.
Scientists are using these tools to mine data for insights that are unreachable with traditional methods, at a scale and speed that were previously unattainable.
“The traditional drug development process is linear and sequential,” says Casper Hoogenraad, Vice President and Head of Neuroscience in Genentech’s Research and Early Development (gRED) organization. “Researchers start with a single target that, based on disease biology or human genetics, is dysregulated and then figure out what kind of therapeutic might modulate the activity of that target be it a small molecule, an RNA approach, or a large molecule, like an antibody.”
“As we continue to push the boundaries of drug discovery, we need new approaches that allow us to ask questions beyond single targets or biological pathways,” says Azad Bonni, Senior Vice President and Global Head of Neuroscience & Rare Diseases, Roche Pharma Research & Early Development (pRED). “We need to understand how numerous potential drug targets work together to drive disease.”
To accomplish this, we need a better, higher throughput, and more parallel way of working. Genentech and Roche are doing just that, striving for the multiplicative benefits of combining advanced computation with innovative research methods. This allows us to ask bigger questions and make immense progress in our understanding of human biology. In turn, this enhanced insight uncovers new therapeutic targets and informs the design and optimization of novel medicines.
Genentech and Roche are currently applying ML across disease areas and therapeutic modalities, with the goal of creating better models for drug discovery that are predictive, generative and interpretable. This trifecta of model characteristics could be used to predict whether a specific molecule can access a target; generate a molecule to bind to that target; and explain how the target and molecule interact with each other.
“For example, ML has become an invaluable tool to discover relationships from cellular profiling data at massive scale,” says Barbara Lueckel, Head of Research Technologies, Roche Pharma Partnering. “And we are also seeing exciting progress in using ML to predict protein structures, eventually bearing the promise for new drug design of complex molecules.”
“We’d like to understand how incredibly complex biological networks misfire or dysregulate in disease and identify the best points to intervene to restore health,” says Tommaso Biancalani, Senior Principal Scientist and Director, AI/ML at Genentech. “AI is already transforming this field, and we are further building this technology to make discoveries we couldn't uncover with traditional methods.”
For example, Genentech scientists in our AI/ML, infectious disease and computational chemistry departments are also using AI to discover new antibiotics. To eliminate bacterial pathogens, antibiotics must penetrate the outer layer (the membrane) of the target. But pathogens have developed ways to keep antibiotics out, and determining which antibiotics can penetrate the membrane can be a laborious process. So, Genentech scientists are using AI technology to examine the chemical structure of billions of potential antibiotics and determine which ones have the potential to bypass the pathogen’s membrane and eliminate it. Then they can synthesize those and test them in the lab.
“Looking at the big picture, when ML is applied in a loop with experiments and data, it bears the potential to impact target and drug discovery in a really powerful way, amplifying many existing efforts at Roche and Genentech,” says Lueckel.
Partnering on a Revolutionary New Approach
Advanced computation is a multifaceted and rapidly evolving field. To supplement our internal efforts and stay at the leading edge of the field, Genentech and Roche have also been engaging with external collaborators.
In December 2021, Roche and Genentech entered into a collaboration with Recursion Pharmaceuticals to explore new territories of cell biology and develop new treatments in key areas of neuroscience and an oncology indication. The partnership will leverage Recursion’s technology-enabled drug discovery platform in combination with our extensive single-cell data generation and ML capabilities to cast a wide, comprehensive net for novel drug targets, and advance and expedite the development of small molecule medicines.
Unlike the conventional approach, which starts with known targets, the partnership will generate and analyze different types of cellular and genetic data – at a huge scale – to build unprecedented maps of human cellular biology. These maps can be leveraged to identify novel biological relationships and ultimately help discover new targets to bring better medicines to patients faster.
“We’re layering a lot of datasets, including high-resolution imaging of how cells respond to genetic changes and chemical perturbations, or disturbances, along with data on how small molecules affect those responses – and using AI to analyze it all,” Bonni says.
Bonni adds that this could be a paradigm shift in how we identify new targets along with therapies to match. This partnership can help identify new medicines that are unattainable using standard methods. It’s an approach that will be particularly useful in neuroscience, a difficult field with a limited number of promising targets and the well-known challenge of getting medicines across the blood-brain barrier.
“The scale of this project is almost unheard of,” Hoogenraad says. “We’ll be screening libraries of small molecules in parallel with genetic perturbation and RNA profiling approaches, so we’ll have an immediate path forward with potential medicines, which is a decisive benefit. There is a lot of risk involved in pursuing novel targets because we just don’t know enough about the underlying biology. Getting more confidence about targets and potential treatments would be a huge leap forward in neuroscience and other disease areas.”
A Strong Commitment
The Recursion collaboration complements other Roche and Genentech partnerships that could improve various aspects of drug discovery and development. In July 2020, Roche and Genentech entered into a collaboration with Reverie Labs to utilize AI for the discovery and development of next-generation kinase inhibitors. In October 2020, Genentech partnered with Genesis Therapeutics to use their deep learning and molecular simulation platform to discover small molecules for challenging targets that would elude other methods. And in August 2021, Genentech acquired Prescient Design, a company with a deep-learning protein design platform and extensive expertise to help identify and design antibodies, with the eventual goal of rapidly designing therapy candidates in silico (on the computer).
Scientists at Roche are also seeking novel approaches to the identification of AAV capsids in partnership with Roche subsidiary Spark Therapeutics and Dyno Therapeutics. Today’s gene therapies are delivered using naturally occurring viruses, which can carry limited payloads and only target certain tissue types. With Dyno’s AI-powered CapsidMap technology, the partners aim to optimize tissue targeting and immune-evading properties, in addition to improving packaging capacity and manufacturability of gene therapy solutions for CNS and liver diseases.
These partnerships, among others, combined with our internal research efforts, exemplify Roche and Genentech’s commitment to advanced computation, and our firm belief that the digitization of drug discovery and development has real potential to make a meaningful difference for patients.
The drug discovery field is at a turning point. “I am more encouraged than I have ever been,” says Lueckel. “The coming years will further demonstrate for which applications advanced computational approaches like ML live up to their promise, but I am optimistic that these technologies will significantly enhance our efforts to bring new medicines to patients as quickly and efficiently as possible.”