GAO Backs Machine Learning in Drug Development

The Government Accountability Office unveiled a report on Tuesday that identifies challenges and opportunities with the use of machine learning throughout the drug development lifecycle. 

Machine learning is already being used in drug development but challenges remain that hinder its adoption and impact. The 86-page document sets forth six policy options, along with their respective opportunities and considerations, to tackle each of the challenges that it identifies.

Breaking down barriers to the use of machine learning in drug development is intended to reduce the time and cost, which supports greater access. “These improvements could save lives and reduce suffering by getting drugs to patients in need more quickly, and could allow researchers to invest more resources in areas such as rare or orphan diseases,” GAO argues.

Only one out of 10,000 initially tested chemical compounds hits the market, research shows. An estimate of drug development cost puts 2013 at as much as about $2.558 billion per new drug.

The US Food and Drug Administration, meanwhile, has been under mandate to bring down the overall drug development cost. A study published in JAMA earlier this month, however, takes issue with the complexity that resulted from the evolution of FDA’s approval programs. Industry groups support the reorganization of FDA’s Office of New Drugs to facilitate drug development. 

Upon request, GAO conducted an assessment on the use of artificial intelligence-powered technologies in drug development “with an emphasis on foresight and policy implications.” The watchdog assessed use in drug discovery, preclinical research and clinical trials. It conducted interviews with government and industry organizations, among others, for the assessment, too. 

GAO further explains that the AI technologies allow drugmakers to “screen more chemical compounds and zero in on promising drug candidates in less time than the current process.”

Challenges include a lack of quality data, research gaps in biology, chemistry and machine learning, human capital and uncertainty around the regulation of machine learning that could limit investment in the field. GAO flags “a low supply of skilled and interdisciplinary workers.” 

Yet another challenge relates to the barriers around data access and data sharing due to costs associated with machine learning adoption, legal issues and a lack of data sharing incentives. 

More sharing of high-quality data could “help companies identify unsuccessful drug candidates earlier in the development process, conserving resources,” the report states. “For example, cost reductions could occur within each step of drug development or as a new compound moves from one step to another.” GAO did not attempt an economic analysis to quantify cost savings. 

To address the array of challenges, the report pinpoints research, data access, standardization, human capital, regulatory certainty and status quo as its six buckets of policy options. It places emphasis on policy options that do not intervene with current efforts to preserve the status quo. 

GAO recommends promoting data sharing via incentives or mechanisms. Yet it cautions that he uptake of machine learning in drug development could also lead to more cybersecurity risks.

In a June 2019 blog post, BIO similarly stressed the need to address challenges for greater use of AI-based precision medicine. “The most significant challenge is the lack of robust datasets, which can be addressed by broader, more diverse cohorts; more robust collection of genomic data; and a data ecosystem that is both interoperable and shared,” said BIO health policy manager, Jeremy Isenberg. “We need to rethink how to collect and share patient data across the healthcare system—especially regarding clinical trials—and better utilize AI to analyze that data at every touchpoint.” Isenberg also cited human capital as another remaining challenge.