How AI can create a virtual programmable human and revolutionize drug discovery
From idea, to lab, to clinic, to approval — it’s a long and expensive process to bring a drug to market. It also rarely succeeds.
Northeastern University researchers You Wu and Lei Xie are proposing another way: a programmable virtual human that uses artificial intelligence to predict how new drugs will affect not just a targeted gene or protein, but the entire body.
“Basically, we want to use AI and incorporate other techniques to build this kind of human representation where you can test a new compound and see how it works,” says Xie, professor of pharmaceutical sciences at Northeastern. “We want to change the drug discovery paradigm from a one-gene perspective to a systemic view of the human body.”
The proposal, which appeared in the recent edition of the peer-reviewed scientific journal Drug Discovery Today, stems from Wu and Xie’s work integrating AI and other tools to create a new way to develop drugs.
Xie says that the current pharmaceutical development model can take 10 to 15 years to develop a new drug, can cost billions of dollars, and ultimately has a 90% failure rate.
Part of the problem is that early stages of drug testing focus on a new drug’s effect on targeted genes in either in vitro or animal models, Xie says.
But a test tube or mouse is a very different environment than the human body, Xie notes.
And even if the results could be directly translated to humans, testing how a new drug performs on a targeted gene or protein doesn’t necessarily consider how the new drug will interact with other systems in the human body.
“Predicting a perfect protein-drug interaction is not the clinical endpoint just as killing cancer cells in a tumor does not represent a drug’s final destination,” Xie says. “You need to see how this drug interacts with all possible molecules, proteins, DNAs, RNAs in the body and how it’s changed gene expression, protein expressions, all these kinds of things.”
That’s where the proposed programmable virtual human comes in.
Editor’s Picks
The proposal builds upon the concept of a pharmacological digital twin — virtual replicas of humans that use AI to simulate how a drug acts in the human body.
But while digital twins draw conclusions from past drug performance – for example, data from how the drug (or a very similar drug) worked during early clinical stages or after being approved for use — “we want to test a new compound you have never seen before in the human body, without actually testing on humans and violating ethical issues,” Xie says.
By integrating physics-based models of biological, physiological and clinical knowledge and machine-learning models trained on data demonstrating how different human systems work, the proposed virtual human can be programmed to predict how a new drug will interact with the body.
Wu says the proposed tool could be especially effective in discovering drugs for complex conditions, such as Alzheimer’s disease, or neurological disorders that affect systems rather than specific genes or proteins.
“For a lot of diseases, there are multiple, multiple genes working together — especially in complex diseases or rare diseases, or diseases we don’t really have a treatment for yet,” says Wu, a postdoctoral researcher at the Center for Drug Discovery at Northeastern.
A virtual programmable human could also answer questions about a new drug’s side effects, toxicity, effectiveness and other factors long before the clinical phase, potentially increasing the success rates of drugs and saving pharmaceutical companies time and money, Xie says.
Xie says that there remain vast amounts of data to collect, models to integrate, industry and academic collaborators to find and funding to pursue before the proposed tool can be deployed widely. But he adds that the proof-of-concept testing is promising and says “definitely we are in progress toward these goals.”
Ultimately, Xie says a virtual programmable human isn’t just an advance in AI or an enhanced digital twin.
“AI will not make a big difference if we don’t change the current paradigm of drug discovery,” Xie says. “What we propose is a different way for drug discovery, by not just using AI to speed up discovery, but redefining what discovery means.”
“True innovation will come when AI helps us understand how the entire human system responds to treatment,” Xie continues. “We’re building a framework that learns and reasons across all levels of biology.”
