Skip to main content
Premium Trial:

Request an Annual Quote

Zephyr AI Looks to Serve Drug Discovery, Value-Based Care Markets

Premium

CHICAGO – Backed by an $18.5 million seed funding round and some early validation of its technology, bioinformatics startup Zephyr AI can spend at least the rest of the year finding partnerships and ramping up its research rather than chasing revenue.

"We are well capitalized now," said CEO David Morgan. "We're cognizant that we want to have a product that's going to produce revenue in somewhat of the near term, but we're not in a hurry to do that."

Zephyr AI has built a technology platform to aggregate, curate, and analyze multidimensional datasets with artificial intelligence to inform drug discovery and patient care. The company draws its data from public databases covering DNA mutations, copy number variants, gene expression, DNA-protein interactions, drug screening, electronic health records, and medical imaging.

Morgan said that the firm started in cancer and diabetes because there is so much genetic data available there, but there may be opportunities in rare and autoimmune diseases.

A partnership with NexImmune, announced March 17, represents a move into immuno-oncology. Under that deal, the companies will collaborate to identify and validate neoantigens that could lead to new drugs for solid and hematological cancers.

Later in March, Tysons Corner, Virginia-based Zephyr announced that it closed the seed funding round, led by Lerner Group Investments and M-Cor Holdings. Other investors included AME Cloud Ventures, BoxGroup, MedStar Health, Verily, and several individuals.

The firm is now looking to start raising a Series A round by the end of 2022 or in the first quarter of 2023. "That will allow us, I think, to move to the next level," Morgan said, which likely will include setting up an in-house wet lab rather than outsourcing experiments.

Such a round would also give Zephyr the means to pursue mergers and acquisitions starting next year. "Unfortunately, the biotech sector is somewhat depressed today, so there's probably some great opportunities out there for M&A," Morgan said.

For now, though, Zephyr, founded a year ago by investment and incubation firm Red Cell Partners, will apply its new investment to further its goal of supporting precision medicine, particularly in diabetes and cancer care.

Washington D.C.-based Red Cell Partners was started in 2020 by Grant Verstandig, who founded consumer-facing digital health firm Rally Health and later sold it to UnitedHealth Group. Verstandig is executive chairman of Zephyr's board. Other board members include MedStar Health CEO Ken Samet, former Celgene CEO Sol Barer, and former Aetna CEO John Rowe.

According to a Zephyr spokesperson, Red Cell did not invest in the seed round because the round was intended to attract new investors and was already oversubscribed. That firm still holds a majority stake in Zephyr, the spokespeson said.

Red Cell is interested in companies developing machine learning technologies like Zephyr. "When you're talking about … petabytes of disparate datasets and sifting through those to find interrelationships, you need machine learning," Morgan said. "It's impossible to do it otherwise."

Morgan himself, a retired officer in the US Marine Corps, joined Zephyr in January after serving as senior VP at Eurofins for nearly four years. He is also a former executive at Laboratory Corporation of America.

Zephyr AI is involved in drug discovery and what Morgan calls "precision analytics," which he breaks into two categories: supporting clinicians and helping healthcare organizations that have taken on financial risk based on outcomes better manage patient populations.

"It's looking at making better decisions on treatment using personalized medicine and then looking at large patient populations in creating a predictability model," Morgan explained. "That becomes really, really important as we move to value-based care."

He said it is close to impossible to operate in a value-based care framework without understanding when a patient might have an emergent medical episode or when an acute illness might progress to a chronic disease.

For this reason, Zephyr is marketing its precision analytics to large insurance companies and to accountable care organizations (ACOs). Morgan said that Zephyr has also had discussions with pharma companies looking to improve patient medication adherence. There may come some point in the future when the firm works with research institutions, as well.

For its precision analytics, the company is drawing on somatic gene expression data, in addition to longitudinal patient records and socioeconomic data. On the drug discovery side, Zephyr is collecting tumor cell-line data as well as somatic tumor xenograft data to predict sensitivity to various compounds.

All the information Zephyr has ingested so far is from open-source databases.

Two papers published in Oncogene last year by Zephyr AI and Red Cell executives and board members discussed the bioinformatics firm's machine learning technology for both drug discovery and for understanding liver tumor cell lines with varying levels of telomerase reverse transcriptase expression.

Morgan said these were "early, early experiments," but that the computers started to map what he has dubbed "vulnerability networks." Such networks match up known mechanisms in tumor cells to how a drug is supposed to create apoptosis.

"We think that we can better identify patients for trials based on gene expression profiles of different tumors in different stages of cancer. We think that we can identify potential combination therapies or [targets that can lead to] novel molecules," Morgan said. "And we think that there is greater opportunity around maybe repurposing drugs, as well, that have other indications."

Zephyr has trained its software to group cells based on gene expression profiles rather than types of cancer. Morgan said, for example, that a gene expression profile from a stage I prostate cancer may be quite similar to one from a stage IV breast cancer. That kind of knowledge could lead to new insights on how drugs might be repurposed.

On the drug discovery side, Zephyr's target market includes pharma companies large and small that are involved in Phase I or pre-Phase I investigations. Partnerships with contract research organizations are part of the plan, but Morgan said that the firm has not yet had discussions with that sector.

"We're going to have to, at some point, do partnerships with CROs," Morgan said. He indicated that part of the money raised from the seed round will go toward wet-lab experiments rather than in silico research.

"Of the insights that the machine has identified on the drug discovery side, we're going to have to do in vitro experiments to validate those to get them to the next level as a potential candidate for Phase I or pre-Phase I," Morgan said. "So I think CROs become more of a partner with us and less of a customer on the drug discovery side."

The potential market on the predictive analytics side is a broad group of anyone trying to improve predictions related to cancer patient risk in ACOs and other value-based care environments.

"We think all healthcare systems are going to move towards value-based care in future, so we want to be ahead of that and provide them predictive tools so they can understand how to go at risk," Morgan said.

This machine learning can help pharma researchers become more efficient, he said, in some cases by helping them "fail faster, fail more predictably, and fail as cheaply as possible," so they can rule out drug candidates.

Morgan said that the two sides of the business are "somewhat interrelated" because the firm has amassed gene expression, proteomic, transcriptomic, and epigenomic data to train its AI technology. The Zephyr AI system can both inform drug discovery and provide insights to optimize cancer treatments.

"We see how certain drugs are indicated, we see that they are used for certain tumor types, but we know that based on certain genomic profiles, there may be better fits for those drugs," Morgan explained.

He compared the firm's model to the oil industry, where data is oil and the Zephyr technology is the rig. "[Data is] certainly not a commodity, but it's much more available now than it ever was," Morgan said. "By taking those disparate datasets and then using our oil rig, which is our machine learning, [we stand] to gain those insights."

Morgan said that Zephyr differentiates itself from other AI companies by feeding its algorithms drug discovery, cell-line data, and xenograft data that does not bias the machine learning with information on what types of cancer it should expect to see based on the tumor profile. "It's coming up with these insights on its own," he said.

The company is also "hyperconscious of the regulatory environment," Morgan said. While Zephyr's software does not require US Food and Drug Administration review, he said the firm's machine learning explains how targets are identified, theoretically streamlining the regulatory process for drug companies.

Zephyr has built a team of about 35 data scientists, computer engineers, software engineers, computational biologists, and molecular biologists. Morgan said that this team gives Zephyr the ability to "clean," curate, and analyze large amounts of data.

The company is expecting to double its workforce by the end of the year, which he said is typical growth for a startup fresh off a funding round.

About half of the workforce is based in the Washington D.C. area, but like so many other companies in the COVID-19 era, Zephyr has embraced remote working to help it recruit employees from all over, including data scientists from major Silicon Valley entities like Google, Facebook, and Yahoo, according to Morgan.

Zephyr does not currently own any data, but likely will when it gets into in vitro research. Morgan said the company is in the process of building its own xenografts to test the outputs of its machine learning. The firm will also file at least five US patent applications on its algorithms and other technology this year.

"So far, we're finding some really, really interesting vulnerability networks that are suggestive of new compounds or combination therapies," Morgan said.

"It's not our intent to do drug development right now," he said. "It's way, way expensive, so I don't want to get distracted from the important work that we're doing … around machine learning."