Data Quantity, Complexity Drives Use of AI in Drug Discovery and Testing

The quantity of data about medicines, diseases, and biology is growing. So too, are the number of companies that employ artificial intelligence in drug discovery. Most of the low-hanging fruit in drug research has already been picked, and the industry is clamoring to make sense of the new data, according to Jeffrey Lu, CEO and co-founder of Engine Biosciences.

“There’s only one way to do that—use machines to process the complexity,” Lu said.

Lu was one of the speakers featured last week during Xconomy’s Xcelerating Life Sciences San Francisco event. His startup’s technology platform uses AI to uncover gene interactions underlying diseases. The company also uses AI to test therapies that target these interactions, an approach it contends is faster, less expensive, and more precise than conventional drug discovery techniques. The company is discovering drugs for its own internal R&D, as well as for the pipelines of its pharmaceutical partners.

AI and machine learning techniques are also finding applications in clinical trials. San Francisco-based startup Unlearn.AI is developing technology that uses historical clinical trial data to create a virtual version of a real person, called a “digital twins.” The digital version predicts what would happen if the patient had received a placebo. This approach is intended to reduce the number of patients needed to test an experimental drug in a clinical trial, while at the same time increasing its statistical power, according to co-founder and CEO Charles Fisher.

Speaking on the panel, Fisher said that that the duration and expense of clinical development cries out for new approaches that will save on both fronts. He added that new technologies can also reduce the risks faced by patient volunteers.

“We really owe it not only to those patient volunteers, but [also] to all of the patients that are waiting for new treatments, to do this more efficiently,” Fisher said.

Atomwise has been applying its AI technology for drug discovery research since 2012. Excitement about AI is driven in part by the potential to address hard to reach drug targets—the notion of drugging the “undruggable,” said Abraham Heifets, the startup’s CEO and another speaker on the panel. AI is offering a way to unlock biology in ways scientists couldn’t do before, he said.

AI technologies are helping the pharmaceutical industry create a billion molecules a month that can be tested in three to six weeks, Heifets said. But those molecules are useful only if scientists have sophisticated, powerful tools to evaluate them. Tied to all these new molecules are vast amounts of data points, which Heifets characterized as both good and bad. While having more data can help researchers find the answers they seek, bad data can lead them astray.

“We still live in a world of garbage in, garbage out,” Heifets said.

Atomwise puts a great amount of effort into data cleaning and data care. Those efforts ensure that a prediction will be borne out by the experiment, Heifets said. Even tiny errors, such as a misplaced decimal point or a mistyped word, can lead to problems because the algorithms learn the bad data, Fisher said. Because the pharmaceutical industry is heavily regulated, Unlearn’s data processing is done in a way such that every step can be traced back. Those measures ensure there is transparency about the data, Fisher said.

Image: iStock/metamorworks