In the same week that Merck and Amgen revealed expanded alliances with AWS, the bioinformatics startup Culmination Bio revealed that it has received $10 million in funding from the venture arms of those companies, Merck Global Health Innovation Fund and Amgen Ventures. Culmination Bio, a spinoff from Intermountain Health, has developed a vast data lake of de-identified patient records spanning over 40 years.
Dr. Lincoln Nadauld, CEO of the startup, notes that the funding is evidence that the data it has collected can address pharma’s longstanding quest to boost the efficiency of drug discovery and development. Frequently, pharma companies have “access to vast datasets, but those datasets are often not the right kind of data,” Nadauld said. “It’s fragmented, or it’s unstructured, or it’s deficient in some fashion.”
Culmination Bio readies data lake based on four decades of patient health records
Consequently, Culmination Bio is confident that the comprehensive data lake it has assembled, which is longitudinal in nature and disease agnostic, fills an unmet need. “We believe Merck GHI and Amgen are saying, ‘This is a unique asset that we haven’t seen in the market. And we want to accelerate its commercial availability and we want to invest in it for those reasons,’” Nadauld said.
Dr. Lincoln Nadauld
Comprising more than 5 million samples and decades of records, the data set can streamline identifying targeted patient cohorts. Culmination Bio continues to expand the data lake with some 300,000 new biospecimens annually.
The underlying data set is unique in several ways, Nadauld said. One, it captures longitudinal, multimodal data and for scores of patient journeys with matched biospecimens. “What we now have is a multimodal data set that’s inclusive of CPT codes, ICD codes, medications, diagnostic tests, imaging, pathology reports, imaging, etc. And then a biospecimen, which of course is valuable because you can derive additional omics data from that, like DNA data or RNA data,” Nadauld said.
Covering the entire spectrum of human disease
In addition, Culmination Bio’s data set is disease-agnostic. While some vendors have created data sets focused on medical specialities. “We have hundreds of thousands of cancer cases, hundreds of thousands of cardiovascular disease cases, hundreds of thousands of autoimmune disorder cases, etc.” Nadauld said. “Across the entire spectrum of human diseases, one dimension where we’re finding uniqueness.”
“We have not only this exclusive agreement but that we have taken de-identified compliant data and put it in a separate data lake and then built querying functions on top of it. So now we can, in a disease agnostic or in an unbiased way, query across all of that data, which you wouldn’t be able to do in the EMR environment.”
Nadauld further addressed the limitations of traditional EMRs, which offer limited querying capabilities. To implement machine learning algorithms for in-depth data analysis, Culmination Bio exported de-identified EMR data along with multimodal medical data into an independent data lake with advanced querying capabilities. “I think that’s one of our major contributions here,” he said. “So now we can, in a disease agnostic or in an unbiased way, query across all of that data, which you wouldn’t be able to do in the EMR environment.”
Medical institutions pioneering AI in healthcare spinoffs: A growing trend?
Culmination Bio is not the only recent spinoff from a renowned healthcare organization. Another example is Anumana, a startup focused on AI-enabled ECG data analysis, which spun out of the Mayo Clinic — specifically through the incubator nference. Anumana draws on electrophysiological, patient history, and outcome data to enable earlier diagnosis of cardiovascular diseases. The company was profiled in the inaugural episode of AI Meets Life Sci, where Anumana’s chief business officer and Mayo’s Dr. Paul Friedman discussed the impact of neural networks on the early detection and treatment of cardiovascular diseases.
Another example of an AI-focused spinoff from a prominent medical institution is Breathonix from the National University of Singapore, which is developing AI-assisted breath tests for disease detection. In addition, Stanford University researchers are focusing on developing generalizable medical AI, working on AI models that can interpret an array of medical data and assist in diverse medical applications ranging from interpreting imaging studies, lab results, genomics data, and offering recommendations or annotations.
In Anumana’s case, the company claims to have assembled the “largest combined dataset of electrophysiological data, longitudinal patient history, and outcomes in the world” through multi-year, exclusive data partnerships with prominent academic medical centers. The data set includes 11 million patients across more than 20 years. The company has partnered with Pfizer.
From individual patient data to a larger-scale precision medicine initiative
Similarly, Nadauld highlights the gargantuan task of assembling a vast data of healthcare metrics, which typically would involve cobbling data together from multiple institutions. Culmination’s customers are coming to the company with defined questions – requesting, say, 1,000 lung cancer patients who have all received a certain drug or had a certain outcome. ‘And we can pull that together in a way that the data is all consistent and uniform, as opposed to them having to go to 10 different institutions, and get 100 cases from 10 different institutions, sign 10 different agreements, work through 10 different legal processes,’ Nadauld said. This approach also mitigates the risk associated with integrating non-uniform data from various sources. ‘And so this is an opportunity for our partners and customers to come and in one stop, get all that they need to inform their R&D efforts.
In his journey from clinical academia to leading precision health initiatives at Intermountain Healthcare, Nadauld has long had an interest in tapping data to transform healthcare outcomes. “While I was at Stanford mapping genomes in cancer patients, I was doing that on a one-off basis — one patient here, another patient there, and we were finding novel therapeutic targets that would change the management of that patient,” he recalled. “And I thought, ‘You know, this works. This precision medicine idea works. I want to apply it to bigger populations.’”
Interactive patient cohort selection tool for precision healthcare research. [Culmination Bio]
Filed Under: clinical trials, Drug Discovery, machine learning and AI
