NEW YORK (GenomeWeb) – Mendel Health, a year-old Bay Area startup, is building a new artificial intelligence platform for use in clinical trials and personalized medicine.
By analyzing electronic medical records and genomic data, Mendel Health believes its AI platform can identify patients for ongoing trials and help clinicians select treatments.
The San Francisco-based company closed a $2 million seed round in July and will seek $10 million more in coming months to develop the platform for these purposes, which should provide Mendel Health with revenue and experience as it expands, according to CEO Karim Galil.
"Genomics isn't worth anything unless you can see how it affects the patient's phenotype," said Galil. "What almost everyone has access to is either the genomic profiling of the patient or some of his or her medical records, but we do not know how this information reflects on the patient's clinical course," he said. "If you can figure out the course of action for all patients flagged with a certain mutation, how they are treated, how they respond to a certain treatment, you can recommend different treatments for them."
Combining genomic data with EMRs is not simple, however. Roughly 70 percent of records are unstructured, meaning they are not machine readable. According to Galil, who is also a physician, institutions like the University of California, San Francisco, or Stanford University employ hundreds to sift through medical records looking for relevant data. "It's a very medically inaccurate process," he said. "And at the moment there is no way that a machine can read that."
Mendel Health's offering, called Mendel.ai, has been created to address that gap. It relies on methods developed in part by CSO Wael Salloum, who previously built AI tools at Columbia University for translating Arabic dialects into English. Salloum was also a senior research scientist at EMR.ai, another Bay Area startup that offers medical dictation software for physicians, before co-founding Mendel Health with Galil.
The result so far is a search engine that can be used to match patients to clinical trials based on their EMRs. While Mendel Health initially pitched Mendel.ai directly to patients, it has since switched to courting hospitals and cancer centers that use it to query their collections of EMRs and genomic data sets to match patients to trials. One early adopter that Mendel Health has disclosed is the Comprehensive Blood & Cancer Center in Bakersfield, California. Galil said that users like the CBCC are eager to adopt Mendel.ai because pharmaceutical companies will pay them per patient enrolled in a trial. Pharma firms benefit because they can more rapidly identify patients that meet their criteria. And Mendel Health also benefits because it gets access to its partners' data, which allows it to refine its AI platform.
"Any time you have clinical trials or you do research, there are rules that define which patients are candidates for the research and other rules that exclude others," said Galil. Hospitals "have the information in the medical records, and the biopsies, but they don't have the tools that show them which patients with which biopsies are eligible for this research. This is what we offer."
While the term 'search engine' brings Google to mind, Mendel.ai's search results are not yet instantaneous. Rather, after generating results, they are checked by Mendel Health's staff of 20 oncologists before they are reported back to the hospital, usually within 48 hours. That way any errors can be corrected and those corrections can be integrated back into the AI so that it learns. Given the time saved to hospitals and pharmas, they are fine with the wait.
"It's still a reasonable turnaround," said Galil. "And every day the AI becomes better."
Gaining access to more data is Mendel Health's priority, Galil noted. The more data it feeds into its platform, the better and more powerful its search results become. This is one of the reasons why companies have not yet been able to solve the challenge of making unstructured EMRs machine readable, as few have had access to enough data to build a platform, Galil said.
That hasn't stopped many from trying though. Deep Six AI, a Pasadena, California-based startup, is also building a deep learning platform to match patients to clinical trials based on unstructured clinical data, including physician notes and pathology and genomic reports. In May, the firm announced a partnership with the Scottsdale, Arizona-based oncology CRO Translational Drug Development (TD2) to use Deep Six AI's platform for clinical trial patient selection. Another firm, called Driver, offers a treatment access platform for cancer patients.
The competitor to beat, though, is IBM Watson Health. In June, for instance, Novartis announced that it was using the Cambridge, Massachusetts-based company's analytical tool to scour its cache of breast cancer data to determine the best courses of action for treating the disease.
Such big data deals are at the heart of the budding AI market, Galil noted.
"If you think of AI, you have to have a lot of data," Galil said. "You invent the right algorithm for that data, and then you package all of that into a product." This is why Mendel Health makes Mendel.ai available to some partners for free, as gaining data access is part of the deal. "The data the hospital shares with us during the course of using our platform is instrumental," said Galil. "We are using that to build our tool."
CGI and AI
Mendel Health scored its biggest data agreement to date in April, when it announced a partnership with Rutherford, New Jersey-based Cancer Genetics (CGI). Under the agreement, CGI gained the right to use Mendel.ai to continuously match patient data with emerging clinical trials based on their EMRs and existing genomic information.
According to CEO Panna Sharma, the company has been impressed with Mendel Health's and considers the firm a "strategic partner." Sharma noted that CGI's relationship with Mendel Health is non-exclusive.
By partnering with CGI, Mendel Health gained access to the longitudinal data for about 150,000 patients. "This is probably one of the biggest datasets out there," said Galil. At the moment, he said, Mendel owns somewhere between 3 and 5 terabytes of medically unstructured data total.
Sharma said the dataset includes genomic and biomarker data shared in a HIPAA-compliant manner, including gene expression, next-generation sequencing, and immunohistochemistry data. Since it partnered with Mendel, CGI has worked to develop a tool called CGI Match Powered by Mendel to match patients with clinical trials, he said. It is currently being made available to alpha users and will undergo a beta launch sometime next year.
"When we started with Mendel, we decided we needed a value proposition that incorporates all markers beyond sequencing, and incorporates medical history, because clinicians will not make the decision to place someone in a clinical trial based solely on a biomarker," Sharma said. "There is a tremendous amount of naivete around that."
As such, the deal has allowed CGI to transition from being seen mainly as a molecular diagnostics company and reference laboratory to being a "full partner" in patient management. Doing that means "going beyond genomics into all the other comprehensive biomarker data and looking at medical history," Sharma noted. CGI Match therefore combines biomarker data with EMRs and uses clinical trials terminology to match patients to trials.
While CGI hopes that CGI Match will make it more attractive to partners compared to offerings from competitors like Foundation Medicine or Neogenomics, Sharma said the industry is likely to benefit as a whole from the type of capabilities that AI can offer.
"This allows us to move the needle to get more people into clinical trials," Sharma said, noting that currently less than 5 percent of cancer patients are enrolled in such trials in the US. By shortening the time and energy it takes to find patients, it will also reduce the cost of those trials, as well as the stress on healthcare providers, and pharma will cover the costs of those treatments. Sharma noted that companies like CGI will also be able to monitor patients in real-time using AI.
"We'll have a better idea if a certain drug is working or not working," Sharma said. "It will give us better, broader epidemiological information about cancer."
Because of these anticipated benefits, Sharma said that CGI is working with a number of companies across the AI landscape, in addition to Mendel Health. "It's an emerging industry and we are still learning our way through it," he said. "I expect that in the next three to five years, it will have a very big impact."
While he described the market as "competitive," Sharma said that no single company will likely dominate, because clients' needs are diverse. "Everyone has a unique value proposition," said Sharma. "Mendel's is its extreme focus on patients and matching patients to clinical trials."