The process of discovering and developing a new drug typically costs $2-4 billion, takes 10-12 years and fails more than 95% of the time either in clinical trials or while going through the approval process. Around half of failures are due to side effects in humans, sometimes discovered after a drug has been fully approved and goes into mainstream use.
Companies around the world with a Chinese connection, however, are using AI to speed up drug development and solve many of the traditional problems the industry has faced.
From quantum physics-based algorithms to miniature cloned hearts, AI is changing drug development by finding new ways to tackle illnesses and then accelerating drug formulation and testing. China will be key to the future of pharmaceuticals, not just for its growing economic might and oceans of data, but its genetic population.
Previously, pharmaceuticals companies have concentrated on their local caucasian client bases. The likes of Tencent and Sequoia China are now putting money into the sector; firms around the world are making their research tools available in Chinese. But there are also China-specific hurdles, both chronic and acute, facing AI drug development. Issues from intellectual property protection to talent acquisition explain why none of the companies interviewed for this piece are fully based in China.
China’s turning point
China has solid experience in producing generic drugs already developed overseas. This is changing, however, as companies are innovating in new techniques and drug discovery. The ongoing reform of the China Food and Drug Administration will accelerate pharmaceutical development in general in China, a country with 114 million people with diabetes and over 700,000 new cases of lung cancer per year, according to McKinsey. There is political backing too. President Xi Jinping is urging the country to make more practical use of AI technology that benefits the real economy.
China could be uniquely placed to take the lead in drug development. Chinese investors are being encouraged to put money into drug and biotech firms abroad to help bring the tech to China. In the first three months of 2018 alone Chinese investors put $1.4 billion into US biotech and drugs firms, compared to $125.5 million in the same period last year.
How do you discover a drug?
With patience and deep pockets.
AI is changing that.
Very roughly, the process involves trying a huge number of similar variants to tackle a problem, such as compounds that target a protein. If any have an effect, they are considered “leads” and then refined. They are tested for safety and enter trials on animals, then humans, and are then submitted for approval.
A miniature heart made by Novoheart for testing drugs for heart conditions, especially those prevalent among ethnic Chinese. (Image source: Novoheart)
AI can be used throughout the process for testing vast numbers of variables—somewhat at odds with the scientific notion of working on a hypothesis—or to model a specific compound.
“Science is very hypothesis-driven. You do things one at a time, modify one variable at a time and hope that eventually you’re going to get to your endpoint,” Ron Li of Novoheart in Hong Kong told TechNode. “This is not mutually exclusive with the AI, big data kind of approach. The big data approach is that if you have a high throughput, even if your starting point is that you’re just fishing… And if you do it enough times you may hit upon something that would not have been possible with the traditional hypothesis-driven approach.”
The health benefits of quantum physics
We never fully know why some companies invest in others, but for the case of Shenzhen and Boston-based XtalPi, speeding up the drug development process with quantum physics, computational chemistry, and AI could save pharmaceutical companies vast amounts of money while making a tidy sum for its own backers.
XtalPi has just announced that Sequoia China led its $15 million Series B with participation from Google and existing investor Tencent. Previous investors include ZhenFund and FreeS Fund and the now total $20 million of investment makes the startup one of the top-funded AI companies in the biotech sector.
How XtalPi utilizes different technologies to predict solid state compounds. (Image credit: XtalPi)
The founders of XtalPi are all Chinese post-docs from MIT who saw the potential for accelerating drug development through computing. But this isn’t just any computing: the compounds used in drugs are fantastically complex. XtalPi uses quantum physics-based algorithms, computational chemistry, and AI to suggest and simulate solid compounds, meaning drugs scientists can accelerate through lengthy rounds of testing.
The name–pronounced “ex tal pie”–is from Xtal used as a shorthand for crystal. Pi because pi occurs in maths, physics, and chemistry. The company identifies with its endless nature.
“We all know that graphite and diamond are both composed of carbon. So they’re chemically identical, but they have hugely different physiochemical properties because the crystal structure inside is very different. The same goes for drug molecules,” Wang Ruyu, director of communications at XtalPi told TechNode in Beijing.
Spaces between molecules mean the same chemical compound can have very different properties, vital in drug efficacy. “We have the ability to simulate the 3D configuration of chemical compounds and predict in a very fast and accurate way how it will perform as a drug candidate,” she added.
“With quantum physics, AI algorithms and computational chemistry, we predict the solid form of chemical drugs—small molecule drugs,” said Wang. The quantum physics approach, expedited by AI, allows XtalPi to predict the toxicity and solubility of the compound, how stable it is, and whether it would have to be refrigerated. This sort of testing has traditionally been empirical, arrived at over thousands of attempts. AI can go straight to the answer.
This approach can be applied throughout the drug development process, not just for initial drug discovery, making it even more effective at speeding up the process and saving firms money.
Wang called the investment by Tencent, Google, and Sequoia “a vote of confidence in our technology.” “Part of [the investment] will go to further R&D, to further empower our algorithm to do more things faster and better. For example, the quantum physics and computational chemistry part of the algorithm generates a lot of high-precision data, and that will be fed into our AI algorithm to build models that can predict more things and give us more capabilities to help pharmaceutical companies with their R&D,” said Wang.
The company uses an array of cloud computing around the world to run its algorithms. Mixing Amazon Web Services (AWS), Tencent Cloud, Google Cloud, and Ali Yun, the firm can deploy a million cores of computing power. XtalPi has been quite the poster child of AWS as it has moved into China.
XtalPi’s main clients are in the US and the West for now, but in recognition of the growing importance of working with Chinese pharmaceuticals, it will look at more ways of moving into China. Two of its founders have already moved back.
Miniature cloned hearts with Chinese characteristics
“If I take a few milliliters of blood from you and then you come back 16-20 weeks later, we’d be able to show you several jars of your own mini hearts beating and contracting in front of you genetically and immunologically identical to your own self,” Ron Li, founder of Novoheart, told TechNode. The firm is developing drug therapies for heart conditions, particularly those common among ethnically Chinese,
Novoheart uses AI to accelerate testing of compounds on miniature cloned heart tissues, both healthy and diseased. A scientist might look at changing one parameter at a time to see what effect a specific drug compound has on heart tissues whereas AI can cope objectively with 20, not only speeding up testing but also increasing the probability of making chance connections that would otherwise not have been tested for.
“We can be very opportunistic,” said Li of the semi-automated process that can run 10,000 drugs in a series of testing.
Novoheart can clone miniature versions of a person’s heart in weeks to test the different effects of drugs. (Image credit: Novoheart)
AI in medicine is nothing new. According to Li, precision medicine (also known as ‘personalized’) and AI in drug research have been underway for 20 years, but applying them to cloning and testing is having a greater impact than previous approaches such as computer modeling of the effect of compounds on tissues, paired with inaccurate animal testing before human trials. Since all the results gleaned from their testing has come from human tissues, Novoheart can sometimes go straight to Phase II trials, skipping Phase I safety testing. Animal trials take up half the time of preclinical testing and cost up to $200 million.
The miniature hearts, around the size of the end of your thumb, can be an exact clone of a patient or used as more broadly representative of an ethnicity or of patient types such as those with arrhythmia and damage from chemotherapy.
“Our business is to be able to develop or co-develop drugs more specifically for conditions and diseases that are more prevalent among the Chinese and Asian populations. Historically, drugs have been designed for Caucasian populations, and there are a lot more diseases over in this part of the world that may have been overlooked,” said Li, who added that the company is focusing even more locally on southern Chinese genetics.
Natural language processing, Chinese edition
Another rich seam in drug discovery is using AI to mine existing scientific research. Huge and growing bodies of research exist which can be valuable to scientists developing new drugs. The sheer size and range of data types have made sifting through it slow, laborious, and expensive.
Running simple keyword searches results in noisy data and incomplete lists. Language context is key—it’s AI that is beginning to make this knowledge accessible. Companies around the world are developing ways to understand research with algorithms and make it available in Chinese too.
“Innovation in the pharmaceutical industry just hasn’t kept up with the pace of information, and AI is the only way to do that,” Dr. Jackie Hunter, CEO of Benevolent AI told TechNode. “We’ve chosen to apply [AI] to one of the hardest problems: drug discovery and development.”
Benevolent AI, based in London, Belgium, and New York, operates an entire drug discovery pipeline boosted by AI. They use algorithms to make sense of the data available to suggest new areas of research and try to identify possible connections between existing knowledge. They apply AI again to accelerate lead optimization when developing compounds, to determine how to run clinical trials of the compounds and then analyze trial data. Using AI has shortened the early stages of drug discovery by 50-60%.
“80% of life science data is unstructured, and manual methods for search and analysis struggle with the growing volume of data,” Jane Z Reed of natural language processing (NPL) company Linguamatics, Cambridge, told TechNode. “Accurate NLP-based text mining identifies the relevant content within the specific context of documents, that has the actual answers to their questions.”
Dr. Hunter, Benevolent AI’s CEO, explained the benefits of using AI to understand the reams of data available:
“It allows you to do things you could never have dreamt of because we can access this huge array of medical data instead of only focusing on a very small amount of data as a normal scientist does. We can make much better decisions. With the technology we can visualize pathways, look at all the genes involved in a disease and visualize them, sort them by pathway or mechanism and then come up with a number of hypotheses for a particular target, using the tech.”
By allowing algorithms to run through so much documentation, they can detect patterns that human scientists might not have considered looking for, never mind reading thousands of pages of text to try to prove. The results can be mapped onto a network of a disease as it is so far understood, which would suggest the developing of different compounds that would target newly-identified nodes of the disease. “It’s a very facile way to do things—you couldn’t ask that sort of question,” said Hunter.
Open Knowledge Map of the term ‘diabetes’. This free-to-use sifts through research on a given topic and sorts it into topics. Try for yourself at Open Knowledge Maps.
AI greatly accelerates lead optimization—choosing and developing the compounds most likely to be effective at tackling a particular issue. Benevolent AI has found that only 10% of the typical number of compounds need to be made and tested, reducing this stage from three years to one. The company has licensed compounds from Johnson & Johnson 2016, one of which will be out of Phase II study by the end of this year, meaning the company has built an AI drug discovery pipeline in two years.
Hunter is certain Benevolent AI will start dealing directly with Chinese companies soon: “The large pharmaceuticals companies have research facilities in China, but China is growing its own pharmaceutical industry and I’m sure that within 5-10 years it will be a force to be reckoned with.”
The Linguamatics system can be fed with documents, scientific papers, patent filings, patient records, and even tweets. The system has already been made operational in Chinese for certain types of searches using vocabularies and terminologies. Additional natural language processing can be added via the API. Companies in China are already producing dictionaries, such as Yitu in Shanghai.
As well as identifying links between things such as specific genes and conditions, the algorithms can more broadly identify which companies are running clinical trials for very specific treatments, who is a key opinion leader on a certain type of medicine as well as types of patient risk. One example is determining which patients are at risk from lung cancer, suggested by nodules on a radiology scan.
“We have many users who have shown 10-fold or more increases in efficiency or similar improvements in speed for many different applications across drug discovery and development,” said Reed.
China market view
Traditionally the West has excelled in pharmaceuticals development. As AI and other technologies become mainstream in the field, however, the balance is shifting. There will be opportunities for cooperation, but competition will be strong, Mark Vermette told TechNode. Vermette is a principal consultant with experience of the impact of AI and machine learning on the drug development industry at Boston-based biotech consulting group Halloran.
“This is a case of ‘coop-etition’. The health problems we’re experiencing are global, and the benefit of drugs and devices being researched in any region are likely to benefit other regions,” said Vermette, “This is a highly competitive market, and China will have a different approach than the rest of the world. AI is highly competitive and shows a lot of value in healthcare and research, so I think collaboration will be primarily between researchers using the technology, not the technology companies themselves.”
As with many aspects of big data handling in China, the current levels of privacy protection could be of benefit to the country’s drug development.
“There are opinions that China’s ability and willingness to aggregate and share patient health data across drug development in China is an advantage,” said Vermette, “Data is a key input to AI for drug development for patient recruitment, outcomes analysis, genotyping, etc. This could be an advantage in drug development but a major challenge to patient privacy, which is a substantial consideration in Europe and the US.”
China’s drug industry is going through significant changes in line with the country’s development and government policy. “When it comes to the pharmaceutical industry, we know that China has been focusing on the production of generic drugs. The market sector is huge here. But we can tell that there are commitments from the Chinese government that they now want to upgrade from the production of generic drugs to their own drugs, or new drugs,” said Novoheart’s Ron Li.
“The good thing is by producing generic drugs, they have the infrastructure, facilities, and scale and with experience. All they have to do now is come up with their own formulations. They need to have IP-protected formulations and then they can go ahead and produce new drugs, and, with the advances in the last decade or so and the returnees, you can see that new drug candidates are starting to emerge. This is a huge market.”
Why (or why not) China?
A shortage of AI talent, medical infrastructure and poor IP protection for new discoveries and techniques are hampering the adoption of the technology within China itself.
Novoheart itself focuses on the Pearl River Delta (PRD) and newly-designated Greater Bay Area across Guangdong. The area has a population the size of France and a GDP exceeding California’s, but Novoheart is based across the border in Hong Kong.
Hong Kong offers better IP protection, more mature financial infrastructure, education and medical infrastructure. “And it’s next to Shenzhen,” said Li, “Shenzhen is a prototyping city so a lot of capital is emerging, but when it comes to education it is not as established as Hong Kong and it does not have its own medical school, which is why we’ve strategically chosen Hong Kong as our location in China and hopefully from here we can expand in to China to collaborate with Shenzhen to work on the PRD.”
Looking at the use of AI in drug development more broadly, it’s still very early days. The sheer length of the discovery, development, testing and approval process means that even with AI accelerating sections of it, the companies we spoke to are still to see their first AI-boosted drugs gain approval. Unlike rapid AI advances in diagnosis, the technology is not going to mean the swift arrival of huge numbers of new wonder drugs any time soon.
AI is currently an additional tool for scientists, but as Benevolent AI’s Jackie Hunter said when asked whether any scientists prefer more traditional research: “We have had some people [like that]… Those people are no longer with us. If you don’t open your mind to doing things differently, what’s the point? You’re not going to harness the power of the technology.”
