Life on Earth would not have existed without proteins. Proteins are made up of amino acids that join to form polypeptide chains. The chains fold in a specific manner that depends on the sequence, location of all amino acids, and their interactions. All these factors determine the three-dimensional structure of a protein, critical for the protein's function. Whether it is the spike glycoprotein of the SARS-CoV-2 or the hemoglobin that carries oxygen in our blood, it is functional due to its structure.

Why is knowing a protein structure important?

Scientists are interested in determining the three-dimensional structures of proteins, as the structure gives insights into how proteins interact with other proteins and macromolecules. The structures also help understand the mechanisms of diseases and enable the effective design of drugs and diagnostics. Knowledge of protein structure helps in crucial applications such as developing proteins and enzymes for industrial applications and vaccines against pathogens.

How did protein structure prediction evolve over the years?

In the mid mid-1950s, Christian Anfinsen was the first to establish a link between a protein's amino acid sequence and its three-dimensional structure, through his work on the enzyme ribonuclease. He proposed a hypothesis which states that a protein's structure can be predicted from its amino acid sequence and the environmental conditions in which the protein folds.

A few years later, John Kendrew and his team elucidated the first 3D structure of a protein— myoglobin, the oxygen-storing protein in muscle cells, in 1958 using X-ray crystallography. Kendrew and Max Perutz jointly received the Nobel Prize in Chemistry in 1962 "for their studies of the structures of globular proteins." Later, in 1972, Anfinsen won the Nobel Prize in Chemistry for his work on ribonuclease.

For a long time, researchers would experimentally determine protein structures using other techniques, such as cryo-electron microscopy and nuclear magnetic resonance (NMR) spectroscopy. However, these techniques required expensive equipment and posed computational challenges in predicting 'lowest free energy conformations,' as proteins can potentially fold into a vast number of conformations. Importantly, these methods required days, months or even years to predict proteins structures. Further, some proteins, such as membrane proteins, are difficult to crystallize, and their structure determination is difficult. As a result, out of ~23 crore protein sequences available, the structures of only about 2,00,000 proteins were known.

Enter AlphaFold. In 2020, Demis Hassabis, an artificial intelligence researcher and John Jumper, a chemist and computer scientist presented an artificial intelligence-based solution to the problem of predicting three-dimensional structures of proteins. Their model—AlphaFold2 by Google DeepMind, was awarded the 2024 Nobel Prize in Chemistry, which they shared with David Baker for his work on computational protein design.

Earlier in 2024, John Jumper delivered a TNQ Distinguished Lecture on “Highly accurate protein structure predictions: Using AI to solve biology problems in minutes instead of years” in Bengaluru and Mumbai, India.

How does AlphaFold2 work?

AlphaFold2 is a tool that uses artificial intelligence and requires the amino acid sequence of a protein as input in the first step of protein structure prediction. AlphaFold2 is trained in recognizing patterns from known protein structures from databases like UniProt or the Protein Data Bank (PDB). The PDB is a database consisting of three-dimensional structures of biological macromolecules, including those of experimentally determined protein structures. In the next step, AlphaFold2 explores which amino acids are likely to interact with each other in the 3D structure. The AI model then uses neural networks to map the relative distances of amino acids. Finally, in an iterative process, AlphaFold2 generates an accurate three-dimensional protein structure.

AlphaFold1—the earlier version and AlphaFold2 both competed in what is known as the "Olympics of Protein Folding"-- the Critical Assessment of Structure Prediction (CASP), a global competition, where researchers predict 3D-structures of proteins purely based on the proteins' amino acid sequences. The participants are given amino acid sequences of proteins whose structures have just been determined experimentally, but these structures are kept secret. AlphaFold2 aced the competition in the 14th edition of CASP in 2020, accurately predicting protein structures and outperforming other methods.

The future with AlphaFold2

Today, AlphaFold2 is available in the open domain; its source code is freely available online to expedite research in the life sciences. During the COVID-19 pandemic, AlphaFold2 accurately predicted the structure of proteins of SARS-CoV-2 and enabled sharing those with researchers globally. This promoted faster vaccine design and accelerated drug development.

AlphaFold2 has nearly solved the major challenges of protein structure prediction, drastically reducing the cost and time required for structure prediction using experimental methods. The researchers of AlphaFold have created a database of more than 200 million protein structures, covering proteins from organisms ranging from viruses to whales. DeepMind has designed this database in partnership with the European Molecular Biology Laboratory-European Bioinformatics Institute (EMBL-EBI).

AlphaFold2 also holds promise in drug discovery for neglected diseases that affect very few people and, hence, do not attract enough funding. Through its accurate structure prediction, the AI model can help discover drugs for diseases such as Chagas disease and Leishmaniasis, bypassing the need for expensive equipment and laboratory work.

These are just a few of the many impactful applications of AlphaFold2. It has dramatically advanced our understanding of protein structures. It is a pioneering breakthrough achieved using an interdisciplinary approach—combining the knowledge of computation, machine learning, chemistry, physics, mathematics, and biology. AlphaFold2 leads the way in which artificial intelligence can help crack the mysteries of life.