China healthcare AI: A learning process

Artificial intelligence (AI) has been on China’s agenda ever since Google’s AlphaGo Master computer program beat Jie Ke, the world’s number one ranked Go player, in a three-match series in 2017. The government wants the core AI industry to be worth RMB150 billion by 2020 and RMB400 billion by 2025. Local authorities think they can get there even faster: the aggregated two-year targets of 19 out 31 provincial governments amount to RMB400 billion.

Healthcare is seen as one of the most promising areas for AI implementation. As of the end of July, there were 126 Chinese start-ups in the space, according to YiOu, a local technology research group. Nearly half focus on medical image recognition and radiology, using computer vision to analyze X-rays and CT scans and assist physicians on their diagnoses. However, some argue the real AI healthcare payoff will come in drug discovery.

“There are a lot more opportunities to optimize drug discovery using AI than in radiology. It takes 10 years and costs an average of $4 billion to take a new drug from discovery to approval – and the failure rate is 90%,” says Michael Keyoung, head of North America at C-Bridge Capital, which focuses exclusively on healthcare. “Reducing costs and shortening the duration of this process, while increasing the success rate in creating better and safer drugs is a bigger problem to solve.”

Most of the AI companies that have been acquired by – or established partnerships with – the largest pharmaceutical players in the US focus on drug discovery. Nevertheless, fewer than 10% of China’s healthcare AI start-ups specialize in this area.

Levels of intelligence

Drug discovery is an attractive prospect for AI application because it offers the opportunity to move into deep learning. Most mainstream AI solutions in China involve a rule-based approach called “human in the loop.” Machines make decisions in very narrow areas based on specific and static rules. Human intervention is required to navigate complicated issues.

Some scientists argue that this doesn’t actually constitute AI because no intelligence is generated: machines don’t understand the questions that they have answered. 

Deep learning, which is a subset of machine learning, takes AI to another level by imitating the workings of the human brain in processing data and creating patterns for use in decision making. Employing this neural network, the machine learns and creates its own rules without specific instruction or human supervision. Just as AlphaGo Master taught itself to win at Go, the healthcare industry wants computers to learn how to discover new drugs.

Recent investment activity in this area includes a $37 million Series B round led by Qiming Venture Partners for Hong Kong-based Insilico Medicine. Baidu Ventures and Sinovation Ventures are among the company’s other backers. 

Nisa Leung, a managing partner at Qiming, notes that the firm’s technology, media and telecom (TMT) team collaborated with the healthcare investment professionals in reviewing the deal. “Our IT team found the technology quite innovative,” she says. “We have many AI portfolio companies at Qiming. We want to make sure that we’re not investing in something repetitive, something that doesn’t really learn or become a human brain.” 

A surprising – yet helpful – insight into how Insilico works is provided by Zao, an app developed by a subsidiary of Chinese photo editing and sharing platform Momo that allows users to insert their likeness into scenes from movies and TV shows. 

Insilico’s primary AI solution is generative adversarial network, or GAN, which comprises two neural networks: a generator and a discriminator. The former creates a fake signal to fool the latter, and the latter spots the difference between real and fake signals. The competitive tension leads to the creation of new signals of unparalleled quality. This is how Zao takes images of an 18-year-old Chinese girl and turns her into Jon Snow on Game of Thrones.

In 2014, Insilico became the first drug developer to use GAN to identify molecules. Over the last three years, the company has been working to get GAN to imagine new molecules with drug-like properties. 

In 2017, GAN was combined with another new AI technology – reinforcement learning – that uses a reward-based system to help machines find the best possible behavior or path in specific situations. Whereas in supervised learning the training data come with an answer key, reinforcement learning offers no answer; the machine learns from experience in deciding what action to take. This was intended to enhance GAN’s imagination capability.

Practical application

As a result of this combination, Insilico recently took 46 days to generate novel molecules that could potentially be used to disrupt the spread of certain cancers and fibrotic conditions. It normally takes up to a year to test hundreds of thousands of molecules and identify one for a specific target protein. The AI-enabled approach involved “imagining” the perfect molecules. The most promising candidates were then synthesized in the lab and was has been successfully tested in mice.  

“After conducting thorough due diligence, we found that a lot of start-ups couldn’t actually do what they claimed to be able to do in AI drug discovery – there’s a lot of bluff out there,” Leung tells AVCJ. “When Insilico’s test results came out, a lot of big pharmaceutical companies expressed an interest in building partnerships to generate novel molecules at a faster pace.” 

Insilico already has a corporate partner in the form of WuXi AppTec, China’s leading pharmaceutical R&D outsourced services provider. It led Insilico’s Series A round in 2018 and the company relocated from Baltimore to Hong Kong shortly afterwards. 

WuXi AppTec could be described as an insider in the business: it tests and validates molecules designed by Insilico. The move to Hong Kong and partnership with WuXi AppTec enables the company to leverage China’s cheaper and faster clinical development capability, and launch more programs. “It is much easier and quicker to recruit appropriate patients in China for clinical trials,” observes Wayne Shiong, a partner at China Growth Capital. 

Insilico has showcased the potential of AI in drug discovery and the company is now looking to fast-track its technology into new treatments. But as for most biotech start-ups in China, it is still very early days.   

 

SIDEBAR: Intelligent imaging  

Two-year-old Shukun Technology has made a name for itself by using artificial intelligence (AI) to help diagnose heart disease. The company, which works with 180 tier-one hospitals in China, claims its algorithm for analyzing medical images can make the diagnosis process up to 10 times faster. China Creation Ventures led a RMB200 million Series B round for the start-up in February.

“One CT session creates some 400 images. We can use them to reconstruct three-dimensional ‘virtual heart’ and give a pre-diagnosis,” says Lei Zhang, Shukun’s chief marketing officer. “It normally takes 30-40 minutes for a doctor to review and analyze all the images. With AI, this process can be cut to four minutes.”

However, there remains a stark difference between theory and practice. A host of legal issues surround the use of AI in this way – all of which come back to the point that doctors hold medical licenses, not technology solutions. 

“If a patient is misdiagnosed, ultimate responsibility for that lies with the doctors; they can be sued for medical malpractice but not the AI,” says Michael Keyoung, head of North America at Asia-focused healthcare investor C-Bridge Capital. “That’s why a physician still goes back and reviews the images and make a proper diagnosis, rather than just reading the AI-generated report.”

This duplication of work has led to complaints from patients that is taking longer to perform diagnoses. As a result, some industry participants question whether AI represents a workable solution in this area. “Diagnosis is not the place to go,” says Ajay Royyuru, a vice president of health care and life sciences research at IBM. “No matter how well you do it with AI, it’s not going to displace the expert practitioner.” 

• Tweet  
• Facebook  
• LinkedIn  
• Google plus  
• Save this article  
• Send to  

• Topics
• Greater China
• Healthcare
• Technology
• Early-stage
• China
• C-Bridge Capital
• Qiming Venture Partners
• China Growth Capital
• artificial intelligence
• Pharmaceuticals