This algorithm designs proteins from scratch to accelerate drug discovery
The proteins that control our lives are like rolling tumbleweeds. Each has a tangled and unique shape, with spiny side branches dotting its surface. Hidden in nooks and crannies are the locks to combat our most notorious enemies – cancer, diabetes, infections or even aging –if we can find the right key.
We just received a universal key maker. In a study published today in Nature, a team led by Dr. David Baker at the University of Washington has developed an algorithm to design tiny protein keys that unlock these targets from scratch. Far from being an ivory tower chase, the algorithm tackled one of the most puzzling drug discovery challenges of our time: can we design drugs based solely on the structure of a protein lock?
They don’t talk about just any drug. Rather than focusing on small molecules, such as Tylenol, the team turned to protein-like molecules called “binders”. Although they may seem exotic, you know them. One example is monoclonal antibodies, which have played a key role in treating severe cases of Covid-19. They are also among our best weapons against cancer. But these therapeutic giants struggle to enter cells, are difficult to manufacture, and are often prohibitively expensive for widespread use.
What about an alternative? Can we tap into the power of modern computing and design similar but smaller and simpler drugs that are just as effective, if not more so?
According to the Baker team study, the answer is yes. By examining almost half a million candidate binding structures for 12 protein targets, the algorithm succeeded in its task, using minimal computational power compared to previous attempts and highlighting the potential results. He also found a “cheat code” that made workbooks more efficient at grabbing their targets.
Here’s the trick: unlike previous tools, the software only needed the structure of the target protein to design binding “keys” from scratch. This is a much simpler approach compared to previous attempts. And because proteins run our internal biological universe, that means new software key makers can help us unlock the secrets of our cells’ molecular life and intervene when they go wrong.
“The ability to generate new proteins that bind tightly and specifically to any molecular target you want is a paradigm shift in drug development and molecular biology more broadly,” Baker said.
Protein binder What?
Our body is governed by a vast consortium of proteins. Like courtesans in a ballroom, each protein bounces around the cell, temporarily grabbing another protein before leaving it to find the next. Specific pairings can launch cellular plots to trigger or inhibit dramatic cellular processes. Some can command a cell to grow or die peacefully. Others can make a cell cancerous or senescent, letting out toxic chemicals and endangering neighboring cells.
In other words, protein pairings are essential for life. They’re also a powerful medicine hack: if a pair sets off a signaling cascade that harms a cell or tissue, we can engineer a “doorstop” molecule to literally break the pairing and stop the disease.
The problem? Imagine trying to separate two intertwined tumbleweeds driving down a highway by throwing a short but flexible stick at them. It seems an impossible task. But the new study presented a recipe for success: the key is to figure out where to separate the two.
Up to the wall
Proteins are often described as beads on crumpled chains into sophisticated 3D structures. That is not exactly correct. The molecular “beads” that make up proteins are more like humanoid robots, with a rigid trunk and flexible limbs called “side chains.”
As a protein assembles, it connects the trunk components of its constituent amino acids into a strong backbone. Like a fuzzy ball of yarn, the frizzies (exposed side chains) coat the surface of the protein. Depending on their position and the backbone, they form pockets that a natural protein partner, or mime, can easily latch onto.
Previous studies have tapped into these pockets to design mimic workbooks. But the process is computationally heavy and often relies on known protein structures, a valuable resource that is not always available. Another approach is to track down “hotspots” on a target protein, but these are not always accessible to binders.
Here, the team tackled the problem in a way analogous to climbers trying to scale a new wall. The climbers are the binders, the wall is the target protein surface. Looking up, there are a lot of holds and toeholds made of side chains and protein pockets. But the bigger ones, the “hotspots”, may not necessarily hold the climber back for the course.
Another approach, the team explained, is to map all catches, even if some appear small. This opens up a new universe of potential binding points – most will fail, but some combinations can be surprisingly successful. A subset of these points are then challenged with thousands of climbers, each trying to identify a promising route. Once the best routes have been cleared, a second round of climbers will explore these routes in detail.
“Following this analogy, we designed a multi-step approach to overcome” previous challenges, the team said.
To begin, the team scanned a library of potential protein backbones and a massive set of side-chain positions that can lock onto a protein target.
The initial sample sizes were huge. Thousands of potential protein backbone ‘trunks’ and nearly a billion possible side chain ‘arms’ emerged for each target.
With the help of Rosetta, the protein structure and function mapping program that Baker’s team developed, the team narrowed the selection down to a handful of promising binders.
The selection of these binders relies on “traditional physics” without exploiting machine learning or deep learning capabilities, said Dr. Lance Stewart, director of strategy and operations at the Institute for Protein Design. , where Baker’s lab is based. This “makes this breakthrough even more impressive”.
The next big question: so that workbooks can link silicone. But do they actually work in cells?
In a proof of concept, the team selected 12 proteins to test their algorithm. Among these were proteins closely implicated in cancer, insulin and aging. Another group focused on fighting pathogens, including influenza surface proteins or SARS-CoV-2, the virus behind Covid-19.
The team screened 15,000 to 100,000 binders for each of the protein targets and tested the best candidates in E.coli bacteria. As expected, the binders were very effective in blocking their targets. Some cut off growth signals that can lead to cancer. Others have targeted a common region of the flu – the flu – which, in theory, could neutralize multiple strains, paving the way for a universal flu vaccine. Even SARS-CoV-2 is not spared, with the “ultra-powerful” binders providing protection against its invasion in mice (these results were previously published).
The study showed that it is possible to design protein-like drugs from scratch. All that is needed is the structure of the target protein.
“The application possibilities seem endless,” remarked Dr Sjors Scheresco-head of structural studies at the MRC Molecular Biology Laboratory in Cambridge, UK, on Twitter, who was not involved in the study.
The algorithm, although powerful, is not perfect. Despite the discovery of millions of potential binders, only a small fraction of designs actually clung to their target. Even the best candidates needed several changes in their amino acid composition for optimal binding to a target.
But it’s groundbreaking work in a field that could fundamentally change medicine. For now, the method and the large data set “provide a starting point” for understanding how proteins interact inside our cells. This data, in turn, could guide even better computational models in a virtuous circle, especially with an added dose of deep learning magic.
This will “further improve the speed and accuracy of the design,” Stewart said. This is “work that is already underway in our laboratories”.
Image credit: Longxing Cao, Brian Coventry, David Baker, UW Medicine