Nobel Prize in Chemistry 2024: Protein Design and Structure Prediction with AlphaFold2

This note covers the Nobel Prize in Chemistry 2024: who won it, what computational protein design and protein structure prediction mean, how proteins fold and why predicting their shape was once called the "50-year-old grand challenge in biology," how David Baker's Rosetta program and Demis Hassabis and John Jumper's AlphaFold2 solved two sides of this problem, how the discovery unfolded, why it matters, and quick facts for exams.

What was the Nobel Prize in Chemistry 2024 awarded for?

The Royal Swedish Academy of Sciences awarded the prize "for computational protein design" to David Baker (one half), and "for protein structure prediction" to Demis Hassabis and John Jumper (one quarter each).

In plain words, the three laureates solved two related puzzles about proteins, the molecules that carry out almost every chemical task inside living cells.

One puzzle was: if you know the chain of building blocks that makes up a protein, can a computer predict the three-dimensional shape it will fold into? The other was the reverse: if you want a protein with a particular shape, can a computer tell you which chain of building blocks will produce it? Baker tackled the second question;

Hassabis and Jumper tackled the first. The official name of this award is the Nobel Prize in Chemistry.

The Nobel Committee for Chemistry, through its chair Heiner Linke, said the discoveries concerned "the construction of spectacular proteins" and "fulfilling a 50-year-old dream: predicting protein structures from their amino acid sequences," adding that both "open up vast possibilities."

Who are the laureates?

David Baker

David Baker was born in 1962 in Seattle, WA, USA. At the time of the award he was a professor at the University of Washington, Seattle, and an investigator at the Howard Hughes Medical Institute, USA.

He received one half of the prize. Baker originally studied philosophy and social science at Harvard, but a textbook on cell biology changed his direction.

From 1993, as a group leader at the University of Washington, he explored how proteins fold and later built a computer program called Rosetta that predicted protein structures from their amino acid sequences; after its success in CASP, his team realised it could also be run in reverse to design brand-new proteins never seen in nature.

Demis Hassabis

Demis Hassabis was born on 27 July 1976 in London, United Kingdom. At the time of the award he was chief executive of Google DeepMind, London, and received one quarter of the prize.

A former chess prodigy who reached master level by age 13, he co-founded DeepMind in 2010 and later led the team that built AlphaFold and AlphaFold2, AI systems that predict protein shapes from their sequences.

John Jumper

John Jumper was born in 1985 in Little Rock, AR, USA. At the time of the award he was a senior research scientist at Google DeepMind, London, and also received one quarter of the prize.

Trained in physics and mathematics, Jumper had earlier worked on simulating protein dynamics with supercomputers before joining DeepMind in 2017, where he co-led the redesign of AlphaFold into AlphaFold2 with Hassabis.

Why does predicting a protein's shape matter so much?

Proteins are built from a limited alphabet of 20 amino acids strung together into long chains.

According to the popular science text, this chain then "twists and folds into a distinct...three-dimensional structure," and that shape is what gives the protein its job: forming muscle fibres, acting as hormones, working as antibodies, or driving chemical reactions as enzymes.

The underlying puzzle is old. In 1958 to 1960, Cambridge scientists John Kendrew and Max Perutz used X-ray crystallography to produce the first three-dimensional protein pictures, work recognised by the Nobel Prize in Chemistry in 1962.

Then American scientist Christian Anfinsen showed in 1961 that a protein always refolds into exactly the same shape, meaning the amino acid sequence alone determines the final structure; he won the Nobel Prize in Chemistry in 1972 for this.

But in 1969 Cyrus Levinthal pointed out a paradox: a protein chain of just 100 amino acids could in theory take on about 10⁴⁷ different shapes.

If folding were a random search through all these possibilities, it would take longer than the age of the universe, yet real proteins fold correctly in milliseconds.

This meant the sequence must somehow encode a fast route to the right shape, and if scientists could decode that logic, they could predict structures directly from sequences, without years of laboratory crystallography work.

The scale of the gap between what experiments could achieve and what researchers needed was enormous. By the time of the award, scientists had experimentally determined roughly 200,000 protein structures through years of painstaking crystallography, yet they had already identified over 200 million protein sequences from studying life on Earth. Each crystallography study could take years of a researcher's working life, and some proteins simply refused to form the crystals the method needs. This gap between known sequences and known shapes is exactly what made the prediction problem, in the words of Nobel Laureate Venki Ramakrishnan quoted in the scientific background, "a 50-year-old grand challenge in biology."

How does AlphaFold2 predict a protein's structure?

From 1994 a competition called CASP (Critical Assessment of Protein Structure Prediction) tested researchers' predictions against real, hidden structures every two years. Progress stayed slow for over two decades: accuracy barely improved.

In 2018 Hassabis entered DeepMind's first AlphaFold model into CASP13 and reached about 60 per cent accuracy, a surprise win, but still short of the roughly 90 per cent accuracy needed to match experimental methods.

The decisive leap came with AlphaFold2, unveiled at CASP14 in 2020. Unlike the first version, it used a type of neural network called a transformer, which can find patterns across huge amounts of data and work out which parts of the input matter most for a given goal.

The press release noted that AlphaFold2 could then predict "virtually all" of the roughly 200 million proteins researchers have identified.

In broad strokes, the process works like this:

1. Feed in the target protein's amino acid sequence.
2. Search large databases for related sequences from many species and known protein structures, building a multiple sequence alignment.
3. Let the network's two main parts, described in the scientific background as the "Evoformer" and the "Structure module," exchange information and refine a map of how far apart each pair of amino acids should sit.
4. Build an actual three-dimensional backbone from this distance information, treating each amino acid's position as a small moveable unit.
5. Refine the structure through repeated passes (recycling) until the predicted shape stabilises.

Draw a flow diagram starting with a protein's amino acid sequence on the left, an arrow into a box labelled "Evoformer" showing a grid of aligned sequences from different species crossing with a grid of pairwise distances, then an arrow into a box labelled "Structure module" producing a folded three-dimensional ribbon shape on the right.

The scientific background explains why this approach worked where earlier attempts had struggled. Before AlphaFold2, researchers had tried comparing many related sequences from different species to spot correlated mutations: pairs of amino acid positions that changed together across evolution, hinting that those positions sit close together in the folded shape. This idea had existed since the 1990s, but the methods could not reliably tell direct contacts apart from indirect ones until around 2016, when CASP12 showed a sudden jump in accuracy using early machine-learning approaches. AlphaFold2 pushed this much further: rather than relying only on pattern-matching, the scientific background argues it effectively learned an underlying picture of the physical forces and typical distances between atoms, and the scientific background links this physical grounding to the model's strong real-world performance.

How does David Baker's computational protein design work?

Baker's question ran the opposite way: given a desired shape, what sequence of amino acids will fold into it? His program, Rosetta, searched a database of known protein fragments for short pieces that resembled parts of the desired target shape, then optimised them using knowledge of how proteins' energy changes as they fold.

The breakthrough example, published in 2003, was a protein named Top7. According to the popular science account, Top7 had 93 amino acids and "almost exactly the structure they had designed," and crucially its shape did not exist anywhere in nature.

Baker explained the philosophy behind this approach with the analogy: "If you want to build an airplane, you don't start by modifying a bird; instead, you understand the first principles of aerodynamics and build flying machines from those principles."

A simplified version of how Baker's team tested a new design is:

1. Choose or draw a target three-dimensional shape that has never existed in nature.
2. Use Rosetta to search for amino acid sequences likely to fold into that shape.
3. Insert the gene for the proposed sequence into bacteria so they manufacture the protein.
4. Determine the resulting protein's actual structure using X-ray crystallography.
5. Compare the real structure with the computer's prediction to check the design worked.

Draw a ribbon diagram showing two coiled alpha-helices sitting beside a sheet made of five zig-zag beta-strands, arranged in a compact ball shape unlike any natural protein fold, labelling it "Top7, 2003."

Before Rosetta, earlier attempts at protein design in the late 1980s and 1990s relied on simple chemical rules rather than full computation. Researchers had noticed that water-repelling (hydrophobic) amino acids tend to sit inside a folded protein, shielded from the surrounding fluid, while water-loving (hydrophilic) ones sit on the surface. Using the rule that hydrophobic amino acids should sit inside and hydrophilic ones outside, scientists such as Regan and DeGrado built a simple four-helix bundle protein in 1988. A later computational design by Dahiyat and Mayo in 1997 redesigned the zinc-finger shape into a small protein fragment of only about 28 amino acids that no longer needed to bind zinc. Baker's Top7 went far beyond these efforts: it was a much larger, 93-amino-acid protein with a folding pattern not found in any previously known natural protein, designed and verified with full computational optimisation of both the backbone shape and the placement of every side chain.

How did the discovery unfold?

Why does this discovery matter?

By October 2024, according to the press release, AlphaFold2 had reached more than two million users across 190 countries. Work that previously took years of laboratory crystallography can now often be done in minutes on a computer.

The press release notes this helps researchers "better understand antibiotic resistance and create images of enzymes that can decompose plastic."

On the design side, Baker's methods have been used to create proteins intended for pharmaceuticals, vaccines, nanomaterials and tiny sensors. The scientific background document also mentions later work on de novo enzymes and protein-based inhibitors against viruses.

Open questions remain: the scientific background notes that designing advanced protein functions, not just structures, including catalysis and dynamic behaviour, is still an active area of research, and that AlphaFold2 itself is "not perfect," though it estimates how reliable each prediction is.

How does this connect to what you study?

This prize links directly to the chemistry and biology topics of proteins, amino acids and enzymes taught at school level. Students who learn that proteins are built from 20 amino acids strung together, and that these chains fold into three-dimensional shapes, are looking at exactly the molecules this prize concerns.

Understanding that a protein's function depends on its folded three-dimensional shape, and not just its chemical formula, explains why enzymes are so specific to particular reactions and why a single change in a gene's DNA sequence can disrupt an entire protein's structure and cause disease. The source material notes that structure determines function for hormones, antibodies, structural proteins and enzymes alike.

It also connects to the growing use of artificial intelligence in science. AlphaFold2 shows how neural networks, usually discussed in computer science classes, can be applied to solve a problem in structural biology that had resisted traditional laboratory methods for fifty years. The scientific background document describes this as possibly "the first real scientific breakthrough of artificial intelligence," illustrating how subjects such as biology, chemistry, physics and computing increasingly overlap in real research.

Finally, the method of X-ray crystallography mentioned throughout this story, used by Kendrew, Perutz and later by Baker's team to check their designed proteins, is a technique rooted in physics and connects the study of light, diffraction and atomic structure to biochemistry.

Quick facts for exams

The Nobel Prize in Chemistry 2024 was announced on 9 October 2024 by the Royal Swedish Academy of Sciences.

It was split between David Baker (one half, "for computational protein design") and Demis Hassabis and John Jumper (one quarter each, "for protein structure prediction").

Baker worked at the University of Washington and the Howard Hughes Medical Institute in the USA; Hassabis and Jumper both worked at Google DeepMind in London.

The prize concerned two linked breakthroughs: AlphaFold2, an AI model that predicts a protein's three-dimensional shape from its amino acid sequence, and Rosetta, a program that designs brand-new proteins that do not exist in nature.

The total prize amount was 11 million Swedish kronor, shared according to each laureate's portion.

Note: Source. The prize facts in this note are from the Nobel Prize's official site, nobelprize.org.

Glossary

• Amino acid — one of about 20 small building-block molecules that link together to form a protein chain.
• Protein — a chain of amino acids that folds into a three-dimensional shape to perform a biological function.
• X-ray crystallography — a method of determining a molecule's structure by analysing how X-rays scatter off a crystal of it.
• Levinthal's paradox — the observation that a protein could theoretically take an astronomical number of shapes, yet folds correctly in milliseconds.
• CASP — Critical Assessment of Protein Structure Prediction, a biennial competition testing structure prediction methods since 1994.
• De novo design — creating an entirely new protein structure that does not exist in nature, rather than modifying an existing one.
• Rosetta — David Baker's computer program for predicting and designing protein structures from amino acid sequences.
• AlphaFold / AlphaFold2 — AI models built by Demis Hassabis, John Jumper and colleagues at DeepMind to predict protein structures from sequences.
• Transformer — a type of neural network architecture that learns which parts of input data matter most for a task.
• Multiple sequence alignment — a comparison of many related protein sequences from different species, used to spot patterns.
• Top7 — the first entirely new protein structure designed by computer, published by Baker's group in 2003.
• Enzyme — a protein that speeds up a specific chemical reaction in a living cell.

Common errors and misconceptions

• Misconception: AlphaFold2 and Rosetta solve the same problem. Correct: AlphaFold2 predicts shape from sequence, while Rosetta (used for design) works the other way, finding sequences that will produce a chosen shape.
• Misconception: The Nobel Prize in Chemistry 2024 was shared equally among all three laureates. Correct: David Baker received one half, while Demis Hassabis and John Jumper each received one quarter.
• Misconception: AlphaFold2 was the first AI protein prediction tool from this team. Correct: An earlier model, AlphaFold (AlphaFold1), was entered at CASP13 in 2018 before AlphaFold2 achieved the real breakthrough in 2020.
• Misconception: Proteins are made from an unlimited variety of building blocks. Correct: Proteins are generally built from a fixed set of 20 amino acids, combined in different sequences and lengths.
• Misconception: Determining a protein's shape experimentally is now unnecessary. Correct: Experimental methods such as X-ray crystallography remain important, and AlphaFold2's predictions are not always perfect, so the model itself estimates how reliable each prediction is.
• Misconception: Top7 was a modified version of a natural protein. Correct: Top7's structure did not exist in nature and had no significant sequence similarity to any known natural protein.

Exam-style questions with model answers

Q1. For what work was David Baker awarded his share of the Nobel Prize in Chemistry 2024? [2 marks]

1. David Baker was awarded his share of the prize "for computational protein design."
2. This meant he developed computer methods, through his program Rosetta, to design entirely new proteins that do not exist in nature.

Q2. Name the three laureates of the Nobel Prize in Chemistry 2024 and state their shares. [2 marks]

1. David Baker received one half of the prize.
2. Demis Hassabis and John Jumper each received one quarter of the prize.

Q3. What is Levinthal's paradox, and why is it important for understanding protein folding? [3 marks]

1. Cyrus Levinthal calculated in 1969 that a protein chain of just 100 amino acids could theoretically take on around 10⁴⁷ different three-dimensional shapes.
2. If a protein folded by randomly trying every possible shape, it would take far longer than the age of the universe to find the correct one, yet real proteins fold correctly within milliseconds.
3. This paradox showed that folding must follow a guided, biased pathway rather than random search, and implied that the amino acid sequence itself must contain all the information needed to reach the correct shape quickly, which is what made structure prediction from sequence seem theoretically possible.

Q4. Explain the difference between the two main achievements recognised by the Nobel Prize in Chemistry 2024. [4 marks]

1. The prize recognised two related but opposite problems concerning proteins.
2. Demis Hassabis and John Jumper, through AlphaFold2, solved the problem of predicting a protein's three-dimensional structure starting from its known amino acid sequence, using a transformer-based neural network trained on large sequence and structure databases.
3. David Baker, through his program Rosetta, solved the reverse problem: starting from a desired three-dimensional shape and working out which amino acid sequence would fold into it, allowing him to design completely new proteins such as Top7 in 2003.
4. Together these advances mean scientists can both predict the shape of almost any known protein and design new proteins with chosen shapes and potential functions.

Q5. Describe how AlphaFold2 predicts a protein's structure from its amino acid sequence. [5 marks]

1. AlphaFold2 begins with the target protein's amino acid sequence as input.
2. It searches large databases to build a multiple sequence alignment, comparing the sequence with related sequences from many species, and also draws on known structures in the Protein Data Bank.
3. Its network, built around a transformer architecture, has two linked parts described as the Evoformer and the Structure module; the Evoformer works out probable distances between pairs of amino acids while exchanging information with the sequence alignment data.
4. The Structure module then builds an actual three-dimensional arrangement of the protein's backbone using this distance information, treating each amino acid's position as a moveable unit that is adjusted step by step.
5. The whole process is repeated, or "recycled," several times to refine the structure until it stabilises into the final predicted shape, which for most proteins now reaches an accuracy close to that of experimental methods such as X-ray crystallography.

Q6. Discuss the significance of the Nobel Prize in Chemistry 2024 for scientific research, including the impact and limits of the work recognised. [6 marks]

1. The prize recognised breakthroughs that solved two long-standing problems in biochemistry: predicting protein structure from sequence, and designing new protein sequences for chosen structures.
2. Before these breakthroughs, researchers had experimentally determined only about 200,000 protein structures through methods such as X-ray crystallography, a slow and labour-intensive process, while over 200 million protein sequences had already been identified.
3. AlphaFold2, developed by Demis Hassabis and John Jumper, has now been used to predict the structure of virtually all of these 200 million known proteins, and by October 2024 had been used by more than two million people from 190 countries.
4. Applications mentioned in the sources include a better understanding of antibiotic resistance and the study of enzymes that can decompose plastic.
5. David Baker's computational protein design, beginning with Top7 in 2003, has allowed researchers to create proteins for use as pharmaceuticals, vaccines, nanomaterials and sensors.
6. However, the scientific background notes that designing advanced protein functions such as catalysis and dynamic behaviour remains an active area of research, and that AlphaFold2's predictions are not perfect, with the model itself providing an estimate of how reliable each prediction is.

Key takeaways

• The Nobel Prize in Chemistry 2024 went to David Baker, Demis Hassabis and John Jumper for work on proteins.
• David Baker won one half of the prize "for computational protein design."
• Demis Hassabis and John Jumper shared the other half "for protein structure prediction."
• Proteins are chains of about 20 kinds of amino acids that fold into specific three-dimensional shapes.
• Christian Anfinsen showed in 1961 that sequence determines shape, while Cyrus Levinthal's 1969 paradox showed folding cannot be random.
• AlphaFold2, unveiled in 2020, uses a transformer-based neural network to predict structures with near-experimental accuracy.
• Baker's Rosetta program enabled the design of Top7 in 2003, the first entirely new protein structure.
• AlphaFold2 has since been used to predict structures for roughly 200 million known proteins.

Test yourself

When was the Nobel Prize in Chemistry 2024 announced?

It was announced on 9 October 2024 by the Royal Swedish Academy of Sciences.

What share of the prize did David Baker receive?

David Baker received one half of the prize, with the other half shared equally between Demis Hassabis and John Jumper.

What organisation employed both Demis Hassabis and John Jumper at the time of the award?

Both were affiliated with Google DeepMind, based in London, United Kingdom.

What is Top7 and why was it significant?

Top7 was a 93-amino-acid protein designed by David Baker's group in 2003, notable because its structure did not exist anywhere in nature.

What competition tested protein structure prediction methods from 1994 onwards?

CASP, the Critical Assessment of Protein Structure Prediction, a biennial blind-testing competition.

Approximately how many proteins has AlphaFold2 been used to predict structures for?

Virtually all of the roughly 200 million proteins researchers have identified.

What neural network architecture made AlphaFold2 a breakthrough over the first AlphaFold?

A transformer architecture, which learns which parts of the input data are most important for the prediction task.