IEarly morning in October david bakerThe University of Washington (UW) protein biologist received the most anticipated phone call of his scientific career. Halfway around the world, Demis Hassabis and John Jumper from Google DeepMind, an artificial intelligence (AI) company, got the same news. The three scientists were awarded the 2024 Nobel Prize in Chemistry for their “computational studies of protein design and structure.”
Although AlphaFold and the ensuing AI revolution in biology garnered much of the attention, the foundations underlying today’s advances in protein structure and design have been laid over several decades. For those unfamiliar with the details, here is the progressive timeline that led to this monumental achievement.
A timeline of decades of advances in protein design and structural research.
julie davie
1972: Anfinsen presents the protein folding problem
In science, it is often difficult to identify when a scientific problem occurs. But most scientists would agree that biochemists planted the seeds of the protein folding problem in the field of protein biology. Christian Anfinsen In 1972, he received the Nobel Prize in Chemistry for his research on ribonucleases, particularly on the relationship between amino acid sequences and biologically active three-dimensional structures.
Based on his studies of ribonuclease enzymes, Anfinsen proposed that all the information needed to determine a protein’s tertiary structure is encoded in its amino acid sequence. Professor Anfinsen said: “Being able to predict the three-dimensional phenotypic consequences of genetic messages in advance would certainly greatly advance our understanding of cellular organization and the causes and control of abnormalities in such tissues.” said. in him Nobel Prize Lecture.
Thus, the race to solve the protein folding problem continued. Over the next several decades, biologists attempted to reliably predict three-dimensional protein structures from one-dimensional sequences.
“But going from sequence to structure has proven surprisingly difficult for weather prediction in biology, at least in part because even relatively small proteins have a huge number of 3D conformations. Because it can be assumed,” Baker explained in a feature article he contributed. scientist at the turn of the century.

David Baker pioneered the creation of de novo proteins that are better suited to solving modern problems than naturally occurring proteins.
Ian C. Haydon, University of California Protein Design Institute
So in 1994, computational biologists at the University of Maryland john malt and Krzysztof Fidelis To enable scientists to collaborate on this problem, we established the Critical Assessment of Structural Prediction (CASP) competition. Every few years, protein biologists compete to predict the structure of several proteins selected by a committee. The computational model that provided the closest match to the experimental data won.
“Proteins are made from amino acid residues, and amino acid residues are made from atoms. We try to model all the interactions between the atoms and how the atoms fold the protein,” said the contest’s first author. In a previous interview, Baker, who joined us from
1998: Rosetta Program launched
Soon, Baker and his team developed a new computer software, Rosetta, that calculates the energies of different configurations and predicts the best structure at the lowest energy.
“For example, this program, which eliminates unlikely structures containing solvent-exposed hydrophobic residues, intelligently samples the entire protein folding landscape and selects perhaps a million or so possible structures for the lowest energy structure. conformationally,” Baker writes in his book. scientist Feature article.
The Rosetta program had two objectives. This helped predict the structure of proteins, but Baker also applied it to the design of new proteins.
2003: Baker introduces the first De Novo protein
“Instead of predicting what structure a sequence will fold into, not long after the first success with structure prediction, we use these methods to create completely new structures, and then I started thinking that maybe we could look at what happens. The sequence could fold into that,” Baker said in an interview earlier this year.
In 2003, Baker and his team The first de novo proteina 93-amino acid protein called Top7.1 According to Baker, the fact that Top7’s X-ray structure matched their predictions well demonstrated that “modern protein design methodologies can design entirely new proteins with atomic-level precision.”
2008: Scientists gamify protein folding and design
Rosetta literally became famous when Baker and his team launched it. Rosetta@Homean initiative that leverages volunteers’ home computers to supplement computing power requirements. When the volunteers who provided their home computers saw the software at work, some of them gave feedback that they wished the program could suggest what to do next.
So Baker worked with university computer scientists to develop the game. foldit Users could play the game by dragging different parts of the protein across the screen to minimize energy consumption. The less energy you have, the more points you get. It was the perfect balance of work and play. In fact, in 2011 a group of Foldit users solve the structure The identity of the protein that scientists have struggled to decipher for decades.2 Citizen scientists also used games to help design new proteins.3
2018: AlphaFold enters the protein arena

Demis Hassabis led the AlphaFold team to achieve unparalleled speed in protein structure prediction accuracy.
deep mind
Meanwhile, Hassabis, an expert in cognitive neuroscience and co-founder of DeepMind, was also the ace of the game. In 2016, his team applied deep neural network expertise to: Start AlphaGoa powerful program that beat Go, the human champion of board games.4 Soon after, Hassabis turned his attention to the problem of protein folding.
The CASP competition has made incremental progress over the years as scientists test different computational models, but the real breakthrough came in 2018, when Hassabis and his team debuted their AI-based program. It happened at CASP13 in 2017. alpha fold.5 Rather than modeling energy dynamics and calculating structures, the team used existing protein sequences and structures to train AlphaFold using a machine learning approach. After learning rules from thousands of examples, AlphaFold can now apply similar patterns to predict structure from arrays.
2020: AlphaFold2 solves protein folding problems

John Jumper, a key figure at AlphaFold, helped solve the protein folding problem.
deep mind
For the next CASP competition in 2020, Jumper and Hassabis are even stronger with the level-up AlphaFold2. In the new version, Most of the test proteins Accuracy comparable to experimental methods.6 The scale of AlphaFold2’s success was such that Moult and other experts declared: 50 years of protein folding problems Almost resolved.
2024: Baker, Jumper and Hassabis win the Nobel Prize in Chemistry
Over the next few years, DeepMind AlphaFold protein structure databasecurrently contains over 200 million structures. Accessing these protein structures has opened the door to a deeper understanding of their function and potential applications across a variety of fields.
“AlphaFold is already accelerating and enabling large-scale discoveries, including elucidating the structure of the nuclear pore complex, and with the addition of this new structure revealing nearly the entire protein universe, we are We hope that more biological mysteries will be solved.” Eric Topolan expert in cardiology and genomics at the Scripps Research Institute Translational Research Institute. DeepMind Blog Article. Jumper and Hassabis received the 2023 Lasker Award for their work on AlphaFold.
On the protein design side, after initial success with Top7, Baker and his team developed several other de novo proteins over the years. According to Baker, what’s particularly noteworthy these days is that coronavirus vaccine Co-developed with (SKYCovione) neil king This is the first de novo drug approved for human use.7 Baker has several additional projects underway across a variety of application areas, including targeted therapeutics, plastic-degrading enzymes, and carbon dioxide fixation proteins.
“Proteins in nature evolved under the constraints of natural selection. So they solve all the problems associated with natural selection in the evolutionary process. But now we can’t make proteins specifically for 21-year-olds. can.cent-The problem of the century. That’s what’s really exciting about this field,” Baker said.
- Kepnick B et al..New protein designs by citizen scientists. Nature. 2019;570(7761): 390-394.
- Silver D and others.Master Go using deep neural networks and tree search.nature. 2016;529(7587):484-489.
- AW senior and others.Improving protein structure prediction using the potential of deep learning. Nature. 2020;577(7792):706-710.
- Jumper J et al. Highly accurate protein structure prediction using AlphaFold. nature. 2021;596(7873):583-589.
- Walls AC, et al. Induction of strong neutralizing antibody responses by protein nanoparticle vaccines designed for SARS-CoV-2. cell. 2020;183(5):1367-1382.e17.