Multimodal AI for Precision Immunotherapy
& Why Breakout Invested in Noetik
Breakout recently invested in Noetik, an AI-native biotech company using advanced machine learning methods to discover and develop cancer immunotherapies. The Noetik platform unleashes the power of self-supervised machine learning on the company’s massive, proprietary multi-modal data sets from human tumors, driving a deep understanding of cancer biology for the discovery and development of therapeutics.
As covered in an Endpoints News Exclusive last week, Noetik’s $40M Series A raise was led by Polaris Partners. We believe this syndicate — which also includes Khosla Ventures and existing investors DCVC and Zetta Venture Partners — epitomizes the future of how computational therapeutics companies will be built. No longer two sides of the aisle (tech vs biotech), these firms working in collaboration also represents the intersection of what will drive value for Noetik and more broadly in this space — the unrelenting focus on building the best computational platform to identify and develop valuable targets and assets, as well as leveraging the mastery, built over decades, of getting the right therapeutics into clinical trials and through approval. As early-stage specialists at the intersection of technology, biology and chemistry, we couldn’t imagine a company more illustrative of our thesis and are thrilled to deepen our relationship with co-founders Ron Alfa, MD, PhD (CEO) and Jacob Rinaldi, PhD (CSO) as board observers.
Why “One Size Fits All” Doesn’t Work in Oncology
Immuno-oncology (IO) is a white-hot area in human health life science, and for good reason. Harnessing patients’ own immune systems to recognize and destroy cancer cells, these therapeutics can be more effective at reducing long-term remission and safer than traditional treatments like chemotherapy and radiation. They are durable and broadly applicable across cancer types. Plus, many immuno-oncology therapies can be tailored to individual patients, limiting side effects of one-size-fits-all treatments.
IO feels like the golden goose of cancer therapeutics. But drug developers face a big hurdle: understanding and tackling tumor immune subtypes.
Even within a single type of cancer, patients can exhibit different immune subtypes. Identifying reliable biomarkers to predict response to immunotherapy is challenging because different tumor immune subtypes may require different biomarkers. E.g., PD-L1 expression is a biomarker for checkpoint inhibitors, but its presence does not guarantee a response to treatment, particularly in different immune subtypes.
All this complexity means that a treatment effective in one subtype or one patient may not work in another, putting drug developers and physicians at a loss when it comes to predicting what will work. Given the significant technology advancements in biotech in the last two decades, it is somewhat baffling that our approach to oncology remains a guessing game.
Uncovering Biology with Unlabeled Data
Enter: machine learning (ML), a transformative tool reshaping how we approach these challenges. ML comes in various forms — supervised, unsupervised, and self-supervised learning — each suited to different types of data and problems to solve.
For the sake of discussion, let’s say you want to sort a massive collection of pet photos into two groups: dogs and cats.
- Supervised learning categorizes your images (i.e., input data) based on species attributes (i.e., output answers or labels) that you provide. You tell the computer what cats and dogs look like by creating and assigning labels to a set of training data: you’d label some images as dogs and others as cats, and the computer will begin to notice the difference. Not only is this going to be time consuming, but it can also cause problems if some photos of hamsters snuck in.
- Unsupervised learning uncovers patterns, structures, or relationships within the data itself, without any labels provided. You provide the images, and the computer decides what attributes can be used to group them into subsets. The model won’t create labels, but it will stratify the images into groups that you can manually skim through, which might let you catch onto the hamster infiltration. You would still need to test the model on just a portion of your files to start before trusting it to keep chugging.
- Self-supervised learning is a subtype of unsupervised learning that makes use of the structure within the data. The model creates its own labels from the input data (which is freaking awesome) by setting up “tasks” where it predicts one part of the data based on other parts. Basically, self-supervised learning models train themselves. By auto-generating the labels for unlabeled data, this approach converts the unsupervised model to a supervised model without the need for manual labeling — which saves a ton of time and (if done well) resources.
Now, back to biotech…
While using AI to identify dogs and cats on a computer might be fun, relying on manual labor to sift through unlabeled biology including tumor biopsies and cancer research data is outdated and simply untenable. Tumor immunology is simply too complex for humans to solve. Combining the unique strengths of various ML models, researchers are beginning to reveal new biological mechanisms — including new cell types and tumor subtypes.
Noetik Unveils Cancer Biology
Noetik has introduced a disruptive solution to the shortcomings of today’s oncology drug discovery and development process, with its multimodal tumor profiling platform that can precisely define tumor immune subtypes. This represents a giant leap forward in how we understand and approach tumor biology, a critical advancement that unlocks a new generation of precision cancer immunotherapies.
At the heart of Noetik’s platform is the seamless integration of diverse data types from human tissue samples, including histology, genomics, spatial proteomics, and spatial transcriptomics. Self-supervised machine learning models create novel representations of tumor biology, essentially digitizing human tumor biology into high-dimensional image data.
“The tumor immune microenvironment is incredibly complex, and that complexity is layered on the genomic diversity of tumor biology. To understand it, we need tools that can synthesize biology at multiple levels simultaneously to unlock higher order patterns,” explains Dr. Ron Alfa, Noetik co-founder and CEO. “Using self-supervised learning methods, models can learn from a diversity raw data to capture new therapeutically relevant definitions of tumor biology, predict biomarkers, and ultimately the goal is to predict response.”
The ML+IO Dream Team
Advancements in both genomics and computation have brought us to a major inflection point, driving a deeper understanding of cancer biology and enabling huge gains in our ability to design better therapies. At Breakout, we believe using computation to process and decode biology — and accurately model intricate interactions within the tumor microenvironment — will redefine oncology.
But drug development is hard. And building with machine learning is hard. But combining them? The task is daunting, even for exceptional founders.
Our decision to invest in Noetik was driven by two key factors: the recognition that a multi-modal approach to understanding cancer biology is poised to redefine the future of cancer treatment and the exceptional team behind the company, one that epitomizes Breakout’s view that the next generation of talent in intelligent biosciences comfortably lives at the intersections of different disciplines. With their deep technical roots and experience building Recursion and the Parker Institute for Cancer Immunotherapy, Noetik’s leaders have spent their careers building at the forefront of cancer research and development. And as we told the team as we got to know them over the past year and a half, this team is “beautifully impatient,” moving at an unmatched pace that the industry deserves.
CEO and co-founder Ron Alfa, M.D., Ph.D. has built his career in addressing some of the most pressing unmet needs in medicine. As the SVP and Head of Research at Recursion, Ron guided the company from its early stages through to its IPO, driving scientific and portfolio strategy across therapeutics from rare disease to oncology and immunology. Ron is joined by co-founder and CSO Jacob Rinaldi, PhD, who used his neuroscience and computational biology background at Stanford to lead oncology efforts at Recursion and pioneered deep learning applications for cancer vaccines at Genentech. CTO Lacey Padron, PhD, brings a depth of expertise in mathematics, engineering, and biology from her time leading development of advanced data science platforms at the Parker Institute for Cancer Immunotherapy and Nuna, as well as a personal passion for oncology as a cancer survivor. CBO Shaf Virani, MD is a trained neurosurgeon but has spent his career structuring biotech partnerships, spending over a decade in business development at Roche and then as CBO of Recursion, alongside Ron and Jacob.
“Noetik represents our dream team for tackling this problem, and we’re continually impressed by their speed,” says Julia Moore, Co-founder and Managing Partner at Breakout Ventures. “In a very short amount of time, they’ve processed thousands of samples, generated more than 750 terabytes of data, and have trained massively multimodal transformers to learn cancer biology.”
Building the Next Frontier in Oncology
The Breakout team is thrilled to partner with Ron, Jacob, Lacey, and Shaf as they capture the complexity of tumor immune subtypes and take a leadership role in defining the next frontier of oncology. Noetik team — it is a privilege to be “beautifully impatient” alongside you.
This article was edited by Breakout’s Managing Partner and Co-Founder, Julia Moore, and our Director of Community, Susanna Harris.
Want to stay up-to-date on our news and perspectives? Follow us on LinkedIn and join our monthly newsletter mailing list!
