Using AI to design new drugs

Antibiotics are drugs designed to fight bacteria that cause infections in the body. Food poisoning, sepsis, urinary tract infections, and bacterial pneumonia are just a few of the conditions that can be treated with antibiotics. Starting with the discovery of penicillin in 1928, antibiotics are estimated to have saved at least 200 million lives and extended the average human lifespan by 23 years.

Because they are so effective, antibiotics are widely prescribed: their global use has grown by 46% over the past two decades. Bacteria and other germs have adapted in response, resulting in a phenomenon known as antimicrobial resistance, or AMR. Drugs that formerly knocked out bacterial infections no longer work as well – or don’t work at all.

Roughly 700,000 people die each year from drug-resistant diseases. Public health experts estimate that failure to address this problem will cause at least 10 million extra deaths annually by 2050, more than the yearly death toll from cancer. The World Health Organization estimates that rising levels of AMR will force the world to spend an extra US$1.2 trillion annually on public health by 2050.

Drug-makers know they need to invest more in AMR research, but are deterred by high costs and long drug-development cycles. From the initial discovery phase all the way to final approval, it takes an average of US$1 billion and more than 10 years to bring a new drug to market.

Even with such high investment costs, 90% of all clinical drug development fails. A faster, more effective approach is needed.

Enter AI

Discovering new drugs is costly, slow, and prone to failure, partly because researchers must screen hundreds of millions of existing drug molecules to identify the basis for new ones. Artificial Intelligence (AI) could effectively function as a virtual chemist, helping researchers design and identify novel molecules that are likely to interact with drug targets.

By allowing researchers to work faster and more cheaply, AI could shrink R&D costs by up to 70%, while helping scientists discover novel lead compounds in months, rather than years. This would make more potential drug candidates available for clinical trial, lifting the overall success rate in what traditionally has been a hit-or-miss process and increasing the odds that a new chemical compound will eventually become an effective, life-saving drug.

Pharmaceutical industry experts predict that AI will be the emerging technology with the greatest impact on their industry in the years ahead, and they are moving aggressively to capitalize on AI’s potential. Over the past seven years, more than 100 partnerships have been formed between AI vendors and big pharma companies, and the world’s top 10 drug-makers are investing in AI by acquiring new technology or collaborating with outside partners.

One of those partners, Huawei, has created an AI model called the Pangu Drug Molecule Model. Named after a figure in Chinese mythology and developed jointly by Huawei’s cloud business and the Chinese Academy of Sciences, the model has been trained on data from 1.7 billion existing chemical compounds. It uses deep learning to predict which of the compounds are most likely to bind themselves to target proteins that contribute to particular diseases. Once researchers know which compounds are likely to work, they can identify those with the highest chance of success and further optimize them, while minimizing the likelihood that the new drug will have negative side effects.

The X Factor

Dr. Liu Bing, of the Medical School at Jiaotong University in Xi'an, China, is one of the first scientists to use the Huawei Cloud AI-Aided Drug Design Services powered by the Pangu Drug Molecule Model.

Dr. Liu and a team of researchers developed a compound that can treat a number of AMR infections. Drug X, as it is commonly known, targets certain proteins in a way that inhibits the DNA replication of bacteria that cause AMR. Drug X serves as the basis for what is expected to become the world’s first new class of antibiotics with new target proteins in nearly 40 years.

Dr. Liu and his team have used the Pangu Drug Model to screen potentially viable compounds from a database containing billions of molecules. Super-high computing power produced a ten-fold improvement in drug screening efficiency, shortening to just one month the time needed to discover this and other compounds, and reducing R&D costs by 70%.

Drug X has been verified by animal experiments and is undergoing preclinical research for an Investigational New Drug (IND) application, a requirement demonstrating that the drug is reasonably safe for use in clinical trials. Patent applications for Drug X have been filed in multiple countries.

Drug design-as-a-service

As Boston Consulting Group points out, realizing AI’s full value requires a transformation of the drug discovery process, with companies investing in new technology and R&D skills, and exploring new business models.

For example, using the Pangu Drug Model, Huawei has begun selling “AI-aided drug design as-a-service,” the first such SaaS platform in China. The service will soon be available in various regions, starting with Asia Pacific, then the Middle East.

Digitalization is paving the way for a broad array of AI-powered solutions in biomedical and clinical research. Some experts predict that within a decade, AI will play a role in designing almost every new drug brought to market. If even part of AI’s benefits are realized, it will re-shape the economic fundamentals of drug discovery in a way that benefits millions of patients worldwide.