CHICAGO – With a fresh cash infusion of $60 million and a key partnership deal with Mayo Clinic this month, Nference is most certainly in a growth phase.
Nference Founder and Chief Scientific Officer Venky Soundararajan said that the multi-omics analytics company has nearly tripled the size of its team in the last year. The firm employs about 150 scientists, engineers, and physicians at sites including its Cambridge, Massachusetts, headquarters, an outpost in Bangalore, India, at a small office in Toronto, and at its newest location, in Mayo's hometown of Rochester, Minnesota.
Nference is expanding in Toronto this year and is looking to open an office in the San Francisco Bay area within the next couple of months to tap into what Soundararajan called "bilingual" life science and technology talent in both regions. "That will be our next hub of innovation at the convergence of life science and technology," he said.
Mayo prominently featured Nference during its presentation at this month's JP Morgan Healthcare Conference in San Francisco. There, the healthcare system unveiled its Clinical Data Analytics Platform, the first product of its recently launched Mayo Clinic Platform, and introduced Nference as the platform's first collaborator — and exclusive biopharmaceutical analytics partner.
The JP Morgan event came about a week after Nference announced a $60 million Series B financing round, led by Mayo Clinic Ventures. NTT Venture Capital also participated in the round.
The two announcements were closely related. "We made the investment because they were going to be our strategic partner," said Mayo Clinic Ventures Chair Andrew Danielsen.
Mayo Clinic Platform is Mayo's new artificial intelligence-centric partnership with Google to improve healthcare delivery through digital health technologies. It is headed by John Halamka, longtime CIO of Beth Israel Deaconess Medical Center in Boston, who joined Mayo at the beginning of the year after serving as executive director of the Health Technology Exploration Center at Beth Israel Lahey Health.
According to Mayo, the Clinical Data Analytics Platform will focus on identifying novel drug targets and biomarkers, matching patients to proper therapies, and the analysis of "real-world evidence" for label expansion, postmarketing surveillance, and drug repurposing. The partnership with Nference also will inform precision patient care by helping clinicians both inside and outside Mayo make sense of massive stores of data in medical records.
The impetus for the platform came from the hiring of Mayo Clinic CEO Gianrico Farrugia, who took over the top job a year ago, according to Danielson. Mayo sees a little more than 1 million patients a year. Farrugia came in with the thought that 1 million is a drop in the bucket in a nation of 325 million, Danielsen explained.
"How do we use the data we have, the knowledge we have to create products and services digitally that certainly help us deliver medical care better at Mayo Clinic, but also products and services that we can greatly scale up the number of patients that we can positively impact?" Danielsen said.
"Mayo has such an enormous wealth of clinical data that is owned by the patient but is entrusted to Mayo," Danielsen added, noting that patients expect that Mayo not only will use that data to provide the best care possible, but to help future patients.
Mayo has 10 million patient records on hand, as well as 25 million pathology slides in a warehouse.
"We occasionally use those in small sets to discover new drugs and … drug targets, discover new uses for existing drugs, and [develop] new diagnostics to say which patients should receive, which drugs to get the best outcomes," Danielsen said. But much of the data is stuck in silos.
"We could do [our work] much better if we would organize all of our data, ensure that it's deidentified, and then provide access to that dataset to a lot of people under strong vigilance, but with tools like artificial intelligence and augmented intelligence to do it on a faster, more efficient scale," he said.
Nference, founded in 2013, has spent the last seven years building augmented intelligence and deep-learning software for synthesizing biomedical knowledge from scientific, regulatory, and commercial literature to support drug discovery and development, drug life cycle management, and precision medicine.
The vendor applies natural language processing and other extraction techniques to make sense out of unstructured text to build longitudinal phenotypic profiles from electronic health records and clinical reports.
Because the information will be used for research and to inform clinical decisions for current patients, these records are deidentified. Nference is helping Mayo deidentify the data and providing analytics software to facilitate deep querying by researchers and clinicians alike.
Nference and Mayo have engaged a third-party statistical validator — Bradley Malin, director of the Health Information Privacy Laboratory at Vanderbilt University — to certify that the data is properly deidentified and that Mayo's federated data architecture is secure and compliant with HIPAA and other relevant regulations.
Mayo personnel will have free access to the Nference platform to interrogate the deidentified data. Outside academicians will be able to apply for access as well. Pharma companies are the only ones who will be paying for the service, Danielsen said.
"Our goal is — and it's pharma's too — that we dramatically increase the efficiency of drug development. We dramatically decrease the time it takes to complete a clinical trial. We dramatically increase our ability to match the right patient phenotypes with the rate medications for a good clinical outcome," Danielsen said.
First, though, the new Mayo platform will have to show that it can produce those results in both the pharma and clinical realms.
Mayo and Nference are expecting to bring the system online April 30 with about 10 million records of structured data because it is easier to work with, Danielsen said. The goal is to have 2.5 million pieces of
unstructured data online by the end of the year and add another 2.5 million each of the following two years. "The unstructured data is going to take time because it's more complex," Danielsen said.
Danielsen said that the data will never leave Mayo, although "people can bring their analytics to our data and they can query it." Mayo prohibits those who query the data from attempting to reidentify patients or even aggregate results with other information such as geolocation tags or consumer financial records.
Mayo also is "thresholding," the data, according to Danielsen. If a query only turns up a small number of results, the organization will not return the results because the risk of reidentification would be too high.
Soundararajan talked of adding "institutional knowledge" to clinician expertise in diagnosing and developing treatment plans for patients.
"Every patient deserves the right to benefit from that institutional knowledge. Obviously when I am a patient at [Massachusetts General Hospital], I want not just the wisdom of the physician that cares for me, but I want to know what is in all of the minds at that institution," said Soundararajan, a Boston-area resident.
"You get an institutional lens into each deidentified patient journey."
That institutional knowledge can lead to the discovery of important associations between, for example, early-stage diseases, genetic markers, and compounds that might be able to treat the conditions.
Soundararajan gave an example of work that Regeneron Pharmaceuticals has done in identifying PCSK9 as a gene target for cardiovascular conditions, leaning heavily on earlier research by the Human Genome Project and UK Biobank to synthesize genomic and structured phenotypic data from imaging reports and ICD-10 codes.
"Imagine if you could now do that with the lens of institutional knowledge … with the actual context around disease progression, around treatment, and therapeutic outcomes, which are rich in unstructured context and text," he said. Nference is trying to unlock that knowledge, with the help of Mayo.
One area of collaboration is in Mayo Clinic's biobanks, which contains millions of samples. Most have been tied to deidentified patient records but have not been sequenced, according to Soundararajan.
"Imagine if Nference were to [analyze] the whole genome on some samples, single-cell RNA sequencing, mass-spec proteomics, immunochemistry on millions of samples. This represents the true convergence of the genome movement with the phenotype movement," Soundararajan said.
"What Nference and Mayo are beginning to do together is to understand the human genotype-phenotype relationship at a scale here that is likely unprecedented. No one has looked at millions of patient-derived samples to conduct sequencing of various sorts and marry that to phenotypes, not just ICD-10 codes, but the actual context, the actual electronic clinical notes."
This kind of data synthesis is slowly chipping away at the old standby, the hypothesis, leading to a new age of "hypothesis-free learning," according to Soundararajan.
"'I think,' and 'I believe' can be augmented with all of the world's knowledge in its various shapes and sizes," Soundararajan said. "There is an unprecedented opportunity to comb all of that knowledge and structure it in a way in which you can make inference more … dogma-free."
The concept is not necessarily new, but Soundararajan said that it has not been applied on the scale Nference and Mayo are envisioning.
The two partners actually have a relationship going back to 2017, when they formed a joint venture called Qrativ. That venture synthesizes and analyzes genomic and clinical data from publicly available databases as well as from research generated at Mayo to inform drug discovery.
Qrativ and its partners have been focusing on finding new treatments for rare diseases and other conditions perceived as underserved in the market.
In the two-plus years since Qrativ formed, the venture has shown preclinical evidence for repurposing several previously approved ophthalmology drugs for cardiology, according to Soundararajan. That work has led to a Qrativ spinout called Qlaris Bio, which has a pipeline in ophthalmology that is assisting Mayo Clinic in several current clinical trials.
The Qrativ experience has opened doors for several collaborations involving Nference and Mayo. The company has been collaborating with a senior cardiologist to look at early diagnosis of pulmonary hypertension by applying AI to "decode" electrophysiology with the help of unstructured EHR text, Soundararajan said.
Nference also has begun working with Mayo's infectious disease group on looking at new HIV treatments. "These sorts of applications have literally nothing in common other than data science," Soundararajan said.