Abstract
A protein is a chain that folds into a three dimensional shape, and that shape is what lets it do its job in the body. When a protein folds into the wrong shape, or clumps together, the result can be disease. This paper is about the science of protein folding and why it matters to medicine: how shape becomes function, how misfolding drives illnesses such as Alzheimer's and Parkinson's, and how computer simulation of a protein in motion helps researchers find places where a drug might bind. The science is the point here. Work like this is only practical because of the large amount of computing behind it, which many hands contribute. We are one of them. We build computers and servers, so we keep a research cluster of our own working on the problem, though we do not run the studies and we do not interpret the results.
Keywords: protein folding, molecular dynamics, misfolding, drug discovery, structure prediction, distributed simulation
Introduction
Why this matters to medicine
Almost everything a cell does, some protein does the actual work. Proteins carry oxygen, copy DNA, break down food, send signals, and hold tissue together. Each one starts life as a long chain of smaller parts, and then folds up into a compact shape. That shape is not decoration. It is the reason the protein works. A protein that folds correctly fits its partners like a key fits a lock. A protein that folds wrong can lose its job, or worse, take on a harmful one.
Many serious diseases trace back to folding gone wrong. Understanding how a protein reaches its shape, and how it moves once it has, gives medicine a chance to intervene earlier and more precisely. This is slow, patient work, and it sits well upstream of any pill or treatment. But it shapes what becomes possible later.
How to read this paper
This paper is about the science, and the science is the whole of the point. It explains why a protein's shape is its function, what happens when folding fails, and what a computer can add that a still picture cannot. Along the way it notes where the computing that makes this research practical comes from, ourselves among the contributors, but those are passing notes. Read it for the science and you will have the main thing.
Proteins, shape, and disease
Shape is function
A protein is built from amino acids strung together in a specific order. That order is written in a gene. On its own, a bare chain does nothing useful. Within moments of being made, the chain folds. Parts that attract each other pull close, parts that repel push away, and the whole thing settles into a shape that is mostly stable. That final shape has pockets, grooves, and surfaces, and those features are what let the protein grab a specific molecule, speed up a specific reaction, or lock onto a specific partner.
Because the job depends on the shape, small changes to the shape can change the job. A pocket that is slightly too wide may no longer hold what it is meant to hold. A surface that folds inward may hide the very spot the protein needs to expose. This is why a single change in a gene, which changes one link in the chain, can ripple outward into a protein that folds differently and works differently, or not at all.
When folding goes wrong
Figure 1. A protein's shape is its function. When it folds wrong and clumps, the result is tied to some of the hardest diseases in medicine.
Folding can fail in more than one way. Sometimes a protein simply cannot reach its correct shape and is cleared away, leaving too little of it. Sometimes it reaches a wrong shape that is still stable enough to linger. And sometimes misfolded copies stick to one another and pile up into clumps, called aggregates, that the cell struggles to remove.
Several of the most difficult diseases in medicine are tied to this. In Alzheimer's disease, fragments called amyloid beta collect outside cells, and a protein called tau forms tangles inside them. In Parkinson's disease, a protein called alpha synuclein misfolds and aggregates. Huntington's disease and ALS also involve proteins that clump in ways the cell cannot clear. Prion diseases are stranger still: a misfolded protein can coax normal copies of the same protein into misfolding too, so the wrong shape spreads. Cystic fibrosis works from the other direction. There, a protein called CFTR often fails to fold correctly in the first place, so it never reaches the place it is needed and cannot do its job.
The common thread is that shape carries the illness. To understand these diseases, and eventually to treat them, researchers need to understand how these proteins fold, why they sometimes fold wrong, and what holds the wrong shapes together.
Seeing how a protein moves
The limits of a static picture
For decades, learning a protein's shape meant slow laboratory work, growing crystals of the protein or using other careful methods to capture its structure. More recently, computer programs have become very good at predicting the folded shape directly from the sequence of amino acids. AlphaFold, described by Jumper and colleagues in 2021, is the best known of these, and it predicts the static folded shape of many proteins with impressive accuracy. This is a real advance, and it has given researchers structures they did not have before.
But a predicted structure, however accurate, is a single frozen picture. A real protein in the body is not frozen. It jiggles, breathes, opens, and closes. It shifts between shapes. A still image shows you one pose. It does not show you the motion, and much of what matters in medicine lives in the motion.
What simulation reveals
Figure 2. A predicted structure is one still picture. Simulation shows the motion, which is where hidden drug pockets appear.
Molecular dynamics is a way to watch that motion on a computer. Instead of one fixed picture, the computer starts from a structure and calculates how every atom pushes and pulls on its neighbors, then steps forward a tiny slice of time, then does it again, and again. Played out, this produces a moving model of the protein flexing and shifting over time.
Watching the motion shows things a single frame hides. A protein may spend most of its time in one shape but briefly visit others. Some of these visits are rare and short, yet they can matter a great deal. A place where a drug could bind, called a pocket, is sometimes not visible in the resting shape at all. It appears only for a moment when the protein moves a certain way. These are often called cryptic pockets, because they stay hidden until motion reveals them.
This is not a hypothetical benefit. During the recent pandemic, large scale distributed simulation of the coronavirus spike protein found pockets that opened only as the protein moved, pockets that a static view would have missed. That is the kind of opening a drug might one day fit into. It is also why simulation of motion works alongside structure prediction rather than replacing it. Prediction gives you an excellent starting shape. Simulation shows you how that shape lives and moves, and where the openings are.
Why it takes so many computers
Figure 3. Because the useful events are rare, the work is split into many pieces, run across many machines, and pooled. This is where our cluster fits in.
Watching atoms move in this way is expensive, and the reason is the scale of time. The pushes and pulls between atoms happen so fast that the simulation has to step forward in slices of roughly a femtosecond, which is a millionth of a billionth of a second. To simulate even a small stretch of real biological time, the computer must take an enormous number of these tiny steps in order.
That alone would be demanding. The harder problem is that the interesting events are rare. A pocket that opens only occasionally might not appear in any single short run. So researchers do not run one long simulation and hope. They run many simulations, started from many points, and pool the results to build a fuller picture of what the protein does and how often. Many independent runs across many machines add up to far more coverage than one machine could give.
This is the idea behind public protein folding research projects that spread the work across large numbers of computers. Each machine takes a piece, runs its share of the sampling, and sends back results that are combined. The science does not need one enormous computer so much as it needs a lot of computing, applied patiently, for a long time.
That computing has to come from somewhere, and much of it is contributed. Universities, volunteers, and companies keep machines running for the effort and pool what they find. We are one of those contributors. Because we build computers and servers, we run a proactive research cluster of our own, always on, taking its share of the work and returning it to the pool. It is a quiet, steady thing, done alongside everyone else who takes part, and it is part of what turns a method that needs a great deal of computing into one that actually gets done. When a rare state or a hidden pocket shows up a little sooner, the pooled effort we are part of is some of the reason.
Where this has helped medicine
The value of this work is not only theoretical. Understanding a protein's shape, and how it moves, has already changed how some medicines are found, and in a few cases it has led to treatments people take today.
Drugs designed from shape
When researchers can see the exact shape of a protein a disease depends on, they can design a molecule to fit it. Two of the clearest examples are older ones. The protease inhibitors that turned HIV from a death sentence into a manageable condition were designed around the structure of an enzyme the virus needs. The influenza drugs that block the neuraminidase enzyme were designed the same way, by fitting a molecule to a known shape. This approach, called structure-based drug design, rests entirely on knowing the shape, which rests on the folding.
Helping a misfolded protein fold
Cystic fibrosis is one of the strongest examples of folding research reaching patients. In most cases the disease is caused by a protein called CFTR that folds incorrectly and never reaches the surface of the cell where it is needed. Once researchers understood the folding defect, they could search for small molecules that help the protein fold and work, called correctors and potentiators. Combinations of these drugs are now approved treatments that meaningfully improve daily life for many people with cystic fibrosis. The medicine works by addressing the folding problem itself.
Cancer, including women's cancers
Some of the clearest results are in cancer, and several have helped women in particular. Many breast cancers are driven by a protein on the cell surface called HER2, and an antibody built to target that protein, trastuzumab, changed the outlook for people whose tumors carry it. Other breast cancers depend on the estrogen receptor, a protein that hormone therapies such as tamoxifen and the aromatase inhibitors are designed to block. Cervical cancer, once a leading cause of cancer death in women, is now largely preventable with a vaccine made of viral proteins that fold and assemble into empty shells, harmless on their own, that train the immune system to recognize the virus. In each case the medicine was built around a protein and its shape.
New targets from watching motion
Simulating motion adds targets that a static picture would miss. When a protein opens a hidden pocket only as it moves, that pocket becomes a place a future drug might grip. The clearest recent case came during the pandemic, when large-scale distributed simulation of the coronavirus spike protein revealed hidden pockets across the virus, giving researchers new places to aim. This is early, upstream work, and most of these leads are still being followed rather than sitting in a pharmacy. But it shows what the method adds: not only a picture of a protein, but a map of the openings that appear when it moves.
The misfolding diseases are more mixed, and more honest to describe as ongoing. In Alzheimer's disease, the understanding that misfolded amyloid drives part of the illness led to antibody treatments aimed at clearing it. Their benefit so far is modest and still debated, and they are not a cure. We include this to stay balanced. Folding research points the way, but the road from a mechanism to a medicine is long, and for the hardest diseases it is far from finished.
What this does and does not achieve
It would be easy to overstate what simulation delivers, so we will not.
Simulation guides and narrows. It suggests where a pocket might be, which shapes a protein tends to visit, and which ideas are worth testing first. That is valuable, because it saves researchers from testing everything blindly. But a result on a computer is a prediction, not a fact about the body. Every promising lead still has to be checked in the laboratory, in cells and in living systems, and then, far later and only if it holds up, in careful clinical trials with real patients.
Simulation is one early link in a long chain. It does not replace wet laboratory experiments, and it certainly does not replace clinical work. Between a hidden pocket found on a screen and a medicine that helps a person, there are years of work that no amount of computing removes. The honest way to describe it is this. The computing we contribute is real and it helps, but it comes early in a long chain, and it is the researchers, the clinicians, and the patients who do the hard, human work that actually reaches people.
Conclusion
A protein's shape is the reason it works, and when that shape goes wrong, disease can follow. Predicting the folded shape is now something computers do well, but a still picture is not the whole story. Watching the protein move reveals the rare and fleeting states, including hidden pockets where a future drug might bind, and that is why simulating motion helps medicine and why it takes so many computers to do it well. The science is the point, and the people doing it are the ones who matter here. It runs, though, on a great deal of computing, quietly provided by many contributors working together. We are glad to be one of them, and glad that the machines we build turn out to be useful for something like this.
References and Further Reading
- Jumper, J., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature.
- Introductory reading on protein structure and function, and how shape determines a protein's job.
- Reviews of protein misfolding and aggregation in neurodegenerative disease, covering amyloid beta and tau in Alzheimer's, alpha synuclein in Parkinson's, and related conditions in Huntington's disease and ALS.
- Background on prion diseases and the spread of a misfolded shape between copies of the same protein.
- Background on cystic fibrosis and misfolding of the CFTR protein.
- Overviews of molecular dynamics, femtosecond time steps, rare event sampling, and the use of many pooled simulations.
- Reports on cryptic pockets found through large scale distributed simulation, including work on the coronavirus spike protein.
- Structure-based drug design, including HIV protease inhibitors and influenza neuraminidase inhibitors.
- The development of CFTR modulator therapies (correctors and potentiators) for cystic fibrosis.
- Anti-amyloid antibody trials in Alzheimer's disease and the ongoing debate over their benefit.
- Trastuzumab and HER2-targeted therapy in breast cancer.
- Endocrine therapy for breast cancer: tamoxifen and the aromatase inhibitors.
- HPV vaccines built from self-assembling viral proteins, and their effect on cervical cancer.