A Japanese startup is using deep learning technology to realize this dramatic advance in the fight against cancer, one of the top causes of death around the world.
PFDeNA Inc. was established in 2016 as a joint venture between DeNA, a Japanese internet giant, and Preferred Networks, Japan’s leading artificial intelligence startup, to solve complex problems. One such problem is cancer detection.
PFDeNA’s cancer research can be traced back to the vision of one of Japan’s pioneering entrepreneurs. In 1999, Namba Tomoko founded DeNA, a mobile and online services company that had extraordinary success in e-commerce and gaming. Namba stepped down from her role as CEO in 2011 to care for her cancer-stricken husband, but her commitment to fighting the disease inspired DeNA to launch a healthcare business with its own bioscience lab in 2014. Meanwhile, Preferred Networks had been conducting research on cancer screening with National Cancer Center Japan since 2015, but needed a partner with expertise in lab operations and business. The two companies decided to use PFDeNA as a platform for collaboration, which began in 2018.
Led by board members including DeNA President and CEO Moriyasu Isao and Preferred Networks CEO Nishikawa Toru, PFDeNA is harnessing the power of deep learning, an artificial intelligence technique modeled on the brain, as a way to detect cancer as early as possible. To do that, the venture is building computer tools as well as a state-of-the art lab that will be able to find almost undetectable signs of cancer in routine blood samples. This “liquid biopsy” approach contrasts greatly with current methods such as radiographic imagining and tissue biopsies.
“We want to transform healthcare from a sick-care model, in which patients are cared for when they become ill, to one based on preventive medicine,” says Yoneyama Hiroshi, executive officer at DeNA and vice president of PFDeNA. With a background in business development and healthcare, Yoneyama is keenly aware of the challenges faced by the medical care system in Japan.
“There’s a dire need for early-cancer detection, not only in Japan but overseas as well,” Yoneyama says. “There are hurdles in the liquid biopsy field but we believe we can overcome them based on the strengths of our two founding companies.”
Each partner brings a unique skillset to the challenge. Preferred Networks’ specialty is developing cutting-edge AI solutions. DeNA is able to quickly make decisions on large-scale investments based on its long experience in mobile services. It’s also a player in the healthcare business, and has accumulated significant experience in negotiating with medical centers as well as lab operations. In 2014, DeNA began a direct-to-consumer genetic testing service called MYCODE, which can detect predisposition to a variety of illnesses. About 90% of MYCODE users have made lifestyle modifications to protect their health.
Looking for molecular changes
PFDeNA aims to screen for 14 types of cancer, including lung and pancreatic cancer, and estimates the domestic market for such services could be worth about 400 billion yen ($3.8 billion). The startup is working to develop a system that can rapidly detect telltale signs of the 14 cancers with just one blood test. These can include changes in the number of molecules that can indicate the likelihood or presence of cancer.
Prostate-specific antigen (PSA), for instance, is a protein produced by the prostate gland that is used to screen for prostate cancer. Genetic mutations can also suggest whether a patient may be more likely to develop certain kinds of cancer. PFDeNA is examining the expression patterns of extracellular ribonucleic acid (exRNA) including microRNA (miRNA) as a potential screening tool for multiple types of cancer. Many cancer researchers expect that certain changes in these miRNA biomarkers can indicate the presence of cancer in various organs.
“In addition to massive computational resources, high-quality data is indispensable for the high-precision deep learning computations needed to create an accurate screening system,” says Abe Motoki, a bioinformatics engineer at Preferred Networks. Abe is in charge of developing a predictive model using deep learning. He also has access to Preferred Networks’ computational resources including the MN-3 supercomputer, recently ranked as the world’s most energy efficient on the Green500 list.
“With a disease like prostate cancer, we only need to look at the levels of just one biomarker, PSA,” Abe says. “But with we are trying to detect multiple types of cancer by analyzing over a thousand exRNA expression levels, which is way more than humans can possibly handle. That’s why we need technology like deep learning.”
A powerful collaboration
Japan provides an ideal location for medical startups such as PFDeNA, in part because of readily available medical checkups covered by employers and municipalities, as well as a wealth of high-quality medical data. At its lab in Tokyo, PFDeNA is analyzing thousands of blood samples provided, with patient consent, by medical institutions such as National Cancer Center Japan. The company is working with more than 10 medical centers as it works toward its goal of building a rapid-screening system that could be part of annual medical checkups in the future. These partnerships, along with collaborations with industry and academia, form a solid foundation that’s giving PFDeNA the best chance of succeeding in its quest.
The Japanese government has also pivoted to support such efforts. With their universal healthcare system, Japanese tend to focus on treating problems, paying less attention to prevention. This tendency, along with the aging population, has increased demand for medical care. While grappling with these issues, the Japanese government is trying to transform the national healthcare system into one that focuses more on prevention. The state is also backing R&D projects in the field of early disease prediction and intervention through programs such as the Cabinet Office’s Moonshot R&D program.
"The Japanese government is very keen to come up with measures for cancer detection and prevention, so we fit into the context of what it’s doing,” says Yoneyama. “We were able to receive cooperation from more than 10 medical institutions because they’re working on this issue, and it’s now a trend. So Japan, as a government and as a whole, is very much backing this movement and taking leadership in this area.”
While PFDeNA works toward publishing the results of its research in academic journals, it’s consulting with the Pharmaceuticals and Medical Devices Agency, the authority responsible for certifying drugs and medical devices in Japan, in order to streamline approval of its services when they’re ready for the market.
“Japan is an aging society, and early cancer detection is one way in which the burden of healthcare costs can be reduced,” says Ishikura Kiyo, associate director of PFDeNA’s healthcare business. “Liquid biopsies are a hot international topic right now. This service would be the first of its kind in the world and it’s a complex challenge to overcome. It’s a long-term journey but we have already begun.”
Note: All Japanese names in this article are given in the traditional Japanese order, with surname first.