Advancements through AI help usher in new era of molecular science

Deciphering the building blocks of life has long been a painstaking process, but now a quiet revolution in computing power is changing everything – with research at the University of Otago’s Faculty of Biomedical Sciences at its forefront.

Blood cancers. Autoimmune diseases. The complex responses of plants to our fast-warming world.

For scientists, trying to unravel such things often begins with looking at the intricate, microscopic machines we know as proteins.

Now, a quiet revolution, driven by rapid shifts in computing power and data-crunching artificial intelligence (AI), is transforming what we know about these tiny building blocks of life.

For two University of Otago biochemists working at its forefront, Professor Peter Mace and Dr Adam Middleton, it represents an exciting new era of molecular science.

The power of proteins

Found at the heart of every living cell, proteins do everything from triggering important chemical reactions to defending our bodies from invaders.

But to understand the individual roles that each one plays, scientists have had to recreate the unique, three-dimensional shapes that define them – a painstaking process akin to piecing together a jigsaw puzzle with no reference image.

“We’ve known that understanding a protein’s structure was always key to understanding how it worked,” Peter explains.

“But getting that information has been slow, difficult and required very specialised equipment. For decades, it made for a huge bottleneck for the entire field of biology.”

During his PhD at Otago, investigating the growth factors that control sheep fertility, Peter spent three years working on a single protein structure. For Adam, studying specific antifreeze proteins found in polar fish and in grasses during his PhD at Canada’s Queen’s University was a five-year endeavour.

Their work and that of countless other scientist relied on a “shot in the dark” process called X-ray crystallography. In essence, it involved making lots of a specific protein and then trying to grow a crystal of it.

Peter likened it to dangling a string in a salty solution to grow crystals, but with a complex molecule made of hundreds or thousands of atoms. Each time a scientist successfully solved one of these 3D structures, they’d deposit it in a massive, open-source database called the Protein Data Bank.

This data, accumulated through decades of hard work, ultimately provided the foundation for the AI-driven revolution sweeping the field today.

But first came another major technological leap. Around the time Peter and Adam came to Otago, a new method called Cryo-electron microscopy, or Cryo-EM, began to accelerate the pace of discovery.

Instead of trying to grow a crystal, this allowed scientists to look directly at a protein on a grid using a powerful electron microscope.

The impact was profound.

Peter says Cryo-Em is particularly good at visualising membrane proteins, which are a large proportion of drug targets.

The technology’s speed was showcased during the Covid-19 pandemic, when it was used to map the structure of the virus’s spike protein within a month of the genome being sequenced.

“The group who studied the spike proteins worked out how to modify that protein to be more stable,” Peter explains.

“And those stabilising mutations ended up being incorporated into the vaccines we received.”

Now, the landscape has changed again with advent of modern machine-learning.

Tools like AlphaFold – developed by Google’s DeepMind and trained on the very data from the Protein Data Bank – can decode a protein’s 3D structure in mere minutes.

“Instead of that pipeline where we would toil away for three or four years to solve a structure, really, all of our projects are kind of flipped now,” Peter says.

“In about 10 minutes, using that AlphaFold program, we can make a prediction of what that 3D structure might look like.”

For Adam, the AI is an “incredible hypothesis generator”.

It’s allowed his team to quickly test ideas on a computer before even setting foot in the lab, saving a large amount of time and effort – and particularly beneficial for resource-constrained places like New Zealand.

Peter says the new tech is proving especially significant for drug development.

While most modern medicines are designed to interact with a specific protein in the body – either to block its function or to enhance it – knowing that protein’s precise shape is invaluable.

“Proteins are basically like a lock and a key,” Peter says.

“If you can design the right key for that lock, then you might be able to block it from doing its bad function that you don’t necessarily want.”

Adam notes that the revolution was only possible because decades of structural data and millions of genomic sequences were made publicly available.

“It allows us to ask new questions and pursue research that was simply not feasible before.”

Exploring new horizons

Right now, Adam is using AI to investigate a previously uncharacterised molecule linked to autoimmune diseases like type 1 diabetes and lupus.

“I run a machine learning model on it and it’s been quite an exciting thing to be looking at,” he says.

“It's helped direct my research and has already generated some pretty neat preliminary data.

“This is fundamental science that I just wouldn't have been able to look at without these tools.”

Similarly, Peter’s lab is focused on a protein complex that controls how blood cells mature, a process that can be disrupted in diseases like acute myeloid leukaemia.

By using a combination of traditional techniques and AI, his team can now map the intricate structures of this complex to understand what goes wrong.

Their hope is to design small molecules that can inhibit the process, pushing the cancer cells to mature rather than simply renew, Peter says.

“It's one of those ways that gene expression can be disrupted, and we're working with collaborators in Glasgow, Japan, and the US who are all interested in this protein complex.”

The technology’s broad applicability is also opening up completely new avenues of research.

Peter’s lab, for instance, is branching into plant biology, studying a similar protein complex that helps plants respond to light and environmental stress.

“The plants have evolved some more components that are not in the human system, and they seem to work kind of differently,” he says.

“We're trying to understand how their 3D structures function and control gene transcription.”

This research has potentially important implications for understanding how plants will fare with the temperature pressures of climate change.

While the biggest breakthroughs to come from the new technologies may still be a way off, Peter and Adam say we’re likely to see exciting advances in the near-term.

“It’s all moving ridiculously fast – and there’s a kind of fog at the forefront of it,” Peter says.

“Some of the approaches coming out of it might not pan out, but some will prove to stay.”

These might include step-changes in the design of diagnostics, such as in rapid antigen tests that we came to know during the Covid-19 pandemic. Instead of relying on tricky antibodies, they might soon draw on AI to create new, more stable proteins that can be used for things like home testing.

“We could see that happening within the next few years,” Adam says.

“These designed proteins are really easy to work with and are usually pretty solid molecules.”

Peter sees the entire shift in his field as a microcosm of the AI world.

“There’s some really good stuff in there, there's some hype, but I think, on the whole, it’s changed our careers. It changes what we do,” he says.

Despite being based at a relatively small university, located at the bottom of the South Pacific, Peter and Adam feel they’ve been well placed at Otago to benefit.

The University’s access to powerful GPU clusters allows them to perform cutting-edge computational work locally, without needing to rely on overseas servers.

Within the Faculty of Biomedical Sciences, Peter says they have a collegial and collaborative network of researchers to support them.

“It also means our students get trained quite well, in terms of being encouraged to show more initiative, and they end up with a good postgraduate degree where they can solve problems,” he says.

“However, we still need government investment in science infrastructure to develop new technologies: AI won’t answer everything without experiments.”