How AI is aiming at the bad math of drug development

A new drug typically takes more than a decade to develop, at a cost of almost $3 billion. That's because about 90% of experimental medicines fail during the various stages of chemical engineering, or during animal or human trials.

So drugmakers and investors are spending billions of dollars to turbocharge the search for new treatments using artificial intelligence. Scientists are looking to discover breakthrough medicines by rapidly identifying new compounds and modeling complex mechanisms in the body, and by automating what used to be manual processes. So far only a trickle of treatments created with the much-hyped technology have reached the testing stage.

1. Why is drug development so hard?

There are about 10,000 diseases that affect humans (most of them are rare), and the majority lack an effective treatment. Often scientists can only guess at the mechanism that's causing an illness, let alone identify a treatment or cure.

There's conjecture, too, in deciding which therapy to try among many trillions of possible medicines that could be chemically synthesized or made with biologics, which are generally produced with genetically engineered cells. That makes for a process marked by trial and error.

2. How does AI change things?

Pharmaceutical companies are joining forces with technology outfits to process vast quantities of data — some of it newly accessed and digitized with the advent of electronic medical records — to spot promising therapies. They're using deep learning, a type of machine learning where neural networks consisting of many layers are trained on large datasets to recognize patterns and make predictions.

While it's too early to tell if AI will lead to important new medicines, it might at the least identify ineffective drugs sooner, before companies burn through lots of money. For example, AI can better predict if experimental drugs have a common, often damning side effect that inhibits the production of enzymes that break down medicines in the body and prevents them from building to toxic levels.

3. How does it work?

Here are some potentially game-changing applications:

To identify proteins that could play a role in disease, AI can sort through masses of information such as data from tissue or blood samples from patients who have the disease and others who don't. This could help researchers zero in on new compounds likely to target those proteins.

Robots can apply thousands of potential drugs to various types of diseased cells in hundreds of thousands of miniature experiments. AI can be used to process millions of images of the cells — more than humans could ever review on their own — to predict whether experimental therapies appeared to disrupt the disease without harming healthy cells.

Crunching data on existing medicines can uncover ideas to repurpose treatments, combine them to make them more potent or find ways to personalize them based on genetic markers.

4. Can AI really crack the body's code?

Maybe. One of the most talked-about breakthroughs came in 2018 from DeepMind, the artificial intelligence arm of Google's parent Alphabet. Its "AlphaFold" tool beat computational biologists at predicting the shapes of proteins, the basic building blocks of disease. (DeepMind's "AlphaGo" had beaten the world's top professional player of the Asian game of Go just two years earlier.)

It helped untangle one of the most vexing problems in biology — so-called protein folding. Figuring out the shape of proteins helps drug-hunters narrow down the sort of molecules that might interact with them and identify medicines to attack disease.

5. Has AI created new drugs?

A handful of potential drugs identified with the help of AI are in human trials. They include four from Salt Lake City-based Recursion Pharmaceuticals, which is using existing medicines in new ways and raised $436 million in an initial share sale in April 2021.

The medicines target conditions ranging from abnormal blood vessels in the brain and spinal cord to a rare genetic disorder that destroys nerve cells, most commonly in babies. Three more come from Exscientia, a U.K.-based startup that also went public in 2021; they aim to treat obsessive-compulsive disorder and cancerous tumors. AI is already being used in medical diagnostics, with the U.S. Food and Drug Administration approving a number of software tools in recent years to assist with diagnosing breast cancer from mammograms.

6. How much is Big Pharma investing?

In an effort to reverse an industrywide slide in drug discoveries, there's been a rush of tie-ups. More than $7 billion was invested across 300 deals from 2013 to 2018, according to one report.

Roche, a Swiss drugmaker, aims to use AI to double the number of medicines it can produce within 10 years. GlaxoSmithKline hopes to double its drug success rate to 20% by using AI, teaming up with partners including DNA testing provider 23andMe and Cerebras Systems, an upstart chipmaker that provides computer systems to crunch very large data sets.