Friday, August 12, 2022

DeepMind’s AlphaFold Could Be the Future of Science — and AI

Must read

Shreya Christina
Shreya has been with for 3 years, writing copy for client websites, blog posts, EDMs and other mediums to engage readers and encourage action. By collaborating with clients, our SEO manager and the wider team, Shreya seeks to understand an audience before creating memorable, persuasive copy.

Here’s a thought: artificial intelligence – what’s it good for?

That may seem crude, given the sheer amount of energy, investment and hype in the AI ​​space, as well as undeniable evidence of technological advancement. After all, these days AI can beat any human in games ranging from chess to Starcraft (DeepMind’s AlphaZero and AlphaStar); it can write a B-college history essay in seconds with a few prompts (OpenAI’s GPT-3); it can draw illustrations on request of surprising creativity and quality (OpenAI’s DALL-E 2).

For AI proponents like Sam Altman, the CEO of OpenAI, these developments herald an era where “creative AI tools will have the greatest impact on creative workflows since the computer itself,” as he said. tweeted last month. That may turn out to be true. But in the here and now I’m still a little disappointed.

Not by what exactly these AI tools can do. Type a short prompt in DALL-E 2 and come back and say: “a medieval painting where the wifi doesn’t workfeels close to magic. Still, people can write essays and draw illustrations, and while GPT-3 and DALL-E 2 can do those tasks faster, they can’t really do them better. They are superhuman in speed, not quality. (The exception in the above group is DeepMind’s game model, which is truly superhuman – just ask poor reports Go Master Lee Se-dol — but until those AI skills can be used in the much more complex real world, it’s usually an interesting research project.)

So AI can be fascinating and cool and even a little scary, but what it isn’t yet can really play a vital role in solving important problems – something that can be seen in the fact that all these advancements to boost America’s sluggish productivity rates.

That’s why the recent news about AlphaFold, an AI model from DeepMind that can predict the three-dimensional structure of proteins, seems truly monumental — ushering in not just a new era in artificial intelligence, but a new era in useful, important science.

A “big challenge” solved

For decades, molecular biologists have been trying to crack what is known as “the problem of protein folding.”

Proteins are the biological drivers of everything from viruses to humans. They start out as series of chemical compounds before folding into unique 3D shapes. The nature of those shapes — much like the amino acids that make them up — determines what proteins can do and how they can be used.

By predicting what shape a protein will take based on its amino acid sequence, biologists could better understand its function and its relationship to other molecular processes. Drugs are often designed using structural information about proteins, and predicting protein folding could significantly accelerate drug discovery, among other fields of science.

The problem with the protein folding problem, however, is that identifying the final structure of a protein has generally taken scientists years of arduous lab work. What researchers needed was an AI algorithm that could quickly identify the final shape of a protein, much like computer vision systems today can identify human faces with astonishing accuracy. Until a few years ago, the best computational biology approaches for predicting protein folding were: still far below the accuracy scientists can expect from experimental work.

Enter AlphaFold. Another product from DeepMind, the London-based AI company that was bought by Google (which later became Alphabet) in 2014AlphaFold is an AI model designed to predict the three-dimensional structure of proteins. AlphaFold blew away the competition in a biennial challenge to predict protein structure in late 2020, performing nearly as well as gold standard experimental work, but much faster.

AlphaFold predicts protein structures through a deep learning neural network trained on thousands of known proteins and their structures. The model used those known compounds to quickly learn to predict the shape of other proteins, in much the same way that other deep learning models can absorb massive amounts of data – in the case of GPT-3 about 45 terabytes of text data – to predict what comes next.

AlphaFold was recognized by the news Science as Breakthrough of the Year 2021, beating candidates like Covid-19 antiviral pills and the application of CRISPR gene editing in the human body. One expert even wondered if AlphaFold became the first AI to win a Nobel Prize.

“A new era of digital biology”

The breakthroughs kept coming.

Last week, DeepMind announced that researchers from around the world have used AlphaFold to predict the structures of some 200 million proteins from 1 million species, comprising just about every protein known to man. All this data is made available free of charge on a database set up by DeepMind and its partner, the European Bioinformatics Institute of the European Molecular Biology Laboratory.

“Essentially, you can see it covers the entire protein universe,” DeepMind CEO Demis Hassabis said at a news conference last week. “We are at the dawn of a new era of digital biology.”

The database basically works like a Google search for protein structures. Researchers can type a known protein and get back its predicted structure, saving weeks or more of work in the lab. The system is already being used to accelerate drug discoverypartly through a sister company to Alphabet called Isomorphic Laboratories, while other researchers are using AlphaFold to identify enzymes that can break down plastic.

The sheer speed that AlphaFold enables should also help reduce research costs. Kathryn Tunyasuvunakool, a DeepMind researcher, told reporters that AlphaFold only needed about 10 to 20 seconds to make each protein prediction. That could be especially helpful for researchers working on neglected diseases like leishmaniasis and Chagas disease, which are constantly underfunded because they mainly affect the desperately poor.

“AlphaFold is the unique and momentous advancement in life sciences demonstrating the power of AI”, tweeted Eric Topol, the director of the Scripps Research Translational Institute.

AI that is useful — now

It may well be that AI models such as GPT-3 that act in general language will ultimately have more influence than a more limited application such as AlphaFold. Language is still our greatest signal of intelligence and possibly even consciousness – as witnessed by the recent controversy over whether another advanced language model, Google’s LaMDA, had become conscious.

But for all their advancements, such models are still far from that leveland far from being really reliable for regular users. Companies like Apple and Amazon have toiled to develop voice assistant AIs worthy of the name. Such models also struggle with bias and honesty, as Sigal Samuel wrote earlier this year, a problem to be solved through politics rather than technology.

DeepMind’s AlphaFold model is not without risks. As Kelsey Piper wrote about AI and its applications in biology earlier this year, “Any system powerful and accurate enough to identify drugs that are safe for humans is inherently a system that will also be good at identifying drugs that are safe for humans.” are incredibly dangerous to humans.” An AI capable of predicting protein structures could theoretically be used for malicious use by someone looking to develop biological weapons or toxins.

To DeepMind’s credit, it considered the potential dangers of opening its database to the public, in consultation with more than 30 biosafety and ethics experts, and concluded that the benefits, including accelerating of developing effective defenses against biological threats. eventual risks. “The accumulation of human knowledge is just a huge advantage,” Ewen Birney, director of the European Bioinformatics Institute, told reporters during the press conference. “And the entities that can be risky will probably be a very small handful.”

AlphaFold — which DeepMind has said is the most complex AI system it has ever built — is a very effective tool that can do things that humans can’t easily do. In the process, it could make those human biologists even more effective at their jobs. And in the age of Covid, those jobs are more important than ever, as is their new AI assistant.

A version of this story was initially published in the Future Perfect newsletter. Sign up here to subscribe!

More articles


Please enter your comment!
Please enter your name here

Latest article