CALGARY⁠ — Proteins are biological shapeshifters. The clusters of amino acids that form the basis of human tissue are constantly folding like origami to digest food, attack viruses and perform the many other bodily functions that keep us alive. The variety of shapes they can take is virtually infinite, more intricate and greater in number than the human mind can grasp.

But scientists are increasingly using artificial intelligence to crack their code. Using deep learning, researchers can better understand and predict the structure of proteins. The results: new insight into biological life, drastic reductions in the time it takes to discover remedies for disease and other ailments, and a flood of new venture capital investment.

“There may be no category where AI has a larger impact over the long term than in bio[tech],” said Rob Toews, partner at Radical Ventures, a Toronto-based VC firm focused on AI.

Drug-discovery companies attracted $4.1 billion in investment in 2021, according to PitchBook, a 42 per cent increase compared with the previous year. 

Toronto’s Deep Genomics raised US$180 million last year in a Series C round led by SoftBank. Other startups have signed strategic agreements with major pharma companies, including San Francisco-based Genesis Therapeutics’s recent collaboration with Eli Lilly; ProteinQure, a Toronto-based company that uses AI to develop drugs, has partnered with major pharma companies including AstraZeneca. The startups Recursion and Zymergen went public last year amid the tech-stock frenzy, though both companies are now trading well below their debut prices. 

A better understanding of proteins’ structure and movements has been one of the biggest AI-fuelled biotech breakthroughs of recent years, unlocking vast new opportunities in drug discovery. 

For many years, researchers had been targeting proteins as if they were static structures, said Evan Feinberg, CEO of Genesis, whose platform uses AI to seek drug remedies. The company raised US$52 million in Series A and A1 funding in 2020, including investment from Radical Ventures.

But proteins’ propensity to change shape has long confounded researchers, who for years were trying to “cut and paste AI methods from other fields,” like image recognition, and apply them to molecular studies, Feinberg said.

“The problem is, a drug, or a protein … are not two pictures of cats. They’re not a set of German text waiting to be translated to English; they’re a completely different species.” 

Genesis, for its part, has used machine learning to develop digital representations of how compounds interact with one another in a way that reflects the natural world, and applied those representations to biophysical simulations.

Researchers at Alphabet’s DeepMind made a massive leap in the area in 2020 when they unveiled AlphaFold, a deep learning system that predicts proteins’ structure. It lets researchers see proteins in 3D, presenting a potential answer to the persistent “protein-folding problem” and accelerating the potential for drug discovery. “It will change everything,” one biologist told Nature.

Feinberg, who co-founded Genesis alongside Ben Sklaroff, said AI platforms would drastically cut back discovery times for new drugs, giving way to new breakthroughs on an array of diseases and ailments. 

“To discover a novel medicine, the pipeline, it’s a very long and very expensive process,” he said. 

“There’s hundreds of severe disorders, if not more, that have a really clear biological underpinning, where we know the protein in the body that’s causing the problem, but what’s difficult is the chemistry.” 

Michael Houghton, a Nobel Prize-winning researcher based out of the University of Alberta’s Li Ka Shing Applied Virology Institute (LKSAVI), told The Logic in April that the organization has already used sophisticated computer algorithms to develop potential cures for Alzheimer’s disease. 

“We’ve put our computational-science talents to work as well as our chemistry talents, and we’ve come up with small molecule drugs that, in animal models of Alzheimer’s, actually restore memory responses and restore deficits in the memory of these animals,” he said at the time. 

Researchers have used computational methods for drug discovery for years, but the technology is only now beginning to drastically cut the time involved.

According to PitchBook’s 2021 annual report on machine learning, AI can “shorten a drug-development lifecycle from 10 years to as little as three years, if molecular interactions can be accurately predicted in a computational environment.”

Radical Ventures’s Toews cautioned that the AI sector, like other tech spaces, will likely witness retrenchment in the short term as markets begin to turn more bearish. But he said companies with true potential offer value regardless of the macro environment. 

“There is going to be a big correction and re-evaluation of the markets,” he said. “Long term, my perspective is that markets go up and down, and a lot of times in these pullbacks, these downturns, a lot of the best opportunities emerge.”