# AlphaFold Release

AlphaFold is an artificial intelligence program developed by DeepMind that predicts protein structures from amino acid sequences. Its 2020 version achieved breakthrough accuracy in the CASP14 competition, leading to the 2024 Nobel Prize in Chemistry for its creators.

AlphaFold is an artificial intelligence program developed by [DeepMind](https://www.wikiprompt.org/wiki/google-deepmind), a subsidiary of Alphabet, that performs predictions of protein structure. It is designed using [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) techniques and has become a landmark achievement in computational biology. The program's second major version, released in 2020, demonstrated accuracy comparable to experimental methods for most protein targets, transforming the field of structural biology.

The development of AlphaFold was recognized with the 2024 Nobel Prize in Chemistry, awarded jointly to Demis Hassabis and John Jumper for protein structure prediction, alongside David Baker for computational protein design. The program's success highlighted the power of [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) approaches in solving long-standing scientific challenges.

## Background

Proteins consist of chains of amino acids which spontaneously fold to form the three-dimensional structures essential to their biological function. Determining these structures experimentally requires techniques such as X-ray crystallography, cryo-electron microscopy, and nuclear magnetic resonance, all of which are expensive and time-consuming. Over 60 years, these methods identified structures for about 170,000 proteins, while more than 200 million known proteins exist across all life forms.

Computational prediction methods had been developed for decades, with homology modeling achieving accuracy close to experimental techniques in the best cases. The Critical Assessment of Structure Prediction (CASP), launched in 1994, provided a rigorous benchmark. By 2016, the best methods scored only about 40 out of 100 on the global distance test (GDT) for the most difficult proteins. AlphaFold first entered CASP in 2018, using an AI deep learning approach.

## AlphaFold 1 (2018)

AlphaFold 1 was built on work from the 2010s that analyzed large databases of related protein sequences to find correlated changes at different residues. Such correlations suggest physical proximity even when residues are not consecutive in the sequence, allowing estimation of contact maps. AlphaFold 1 extended this by predicting a probability distribution for distances between residues, effectively creating a distance map, and used advanced learning methods for inference.

In December 2018, AlphaFold 1 placed first overall in the 13th CASP competition. It was particularly successful on the most difficult targets, where no existing template structures were available. However, the code was not made publicly available except for running CASP sequences.

## AlphaFold 2 (2020)

AlphaFold 2, which competed in CASP14 in November 2020, represented a significant architectural departure from its predecessor. AlphaFold 1 used separately trained modules combined with a physics-based energy potential. AlphaFold 2 replaced this with interconnected sub-networks forming a single, differentiable, end-to-end model based on pattern recognition. A final refinement step applied local physical constraints using energy minimization based on the AMBER force field.

Key to the system were two modules believed to use a [transformer](https://www.wikiprompt.org/wiki/transformer) design, progressively refining information about relationships between amino acid residues and between positions and sequences in the input alignment. The program was trained on over 170,000 protein structures from the Protein Data Bank using between 100 and 200 GPUs. Training data also included the Big Fantastic Database, containing 65,983,866 protein families represented as multiple sequence alignments and Hidden Markov models, covering over 2.2 billion protein sequences from reference databases, metagenomes, and metatranscriptomes.

At CASP14, AlphaFold 2 scored above 90 on the GDT for approximately two-thirds of proteins, where 100 represents a complete match to experimental structure. This was described as "astounding" and "transformational." However, some researchers noted insufficient accuracy for a third of predictions and that the underlying mechanism of protein folding remained unsolved.

## Publication and Impact

On 15 July 2021, the AlphaFold 2 paper was published in Nature alongside open-source software and a searchable database of species proteomes. As of November 2025, the paper had been cited nearly 43,000 times. The release made protein structure prediction accessible to researchers worldwide, accelerating work in drug discovery, enzyme design, and understanding disease mechanisms.

Hassabis and Jumper had previously won the Breakthrough Prize in Life Sciences and the Albert Lasker Award for Basic Medical Research in 2023 for their leadership of the AlphaFold project. The 2024 Nobel Prize in Chemistry further cemented the achievement's significance.

## AlphaFold 3 and Aftermath

AlphaFold 3 was announced on 8 May 2024, extending predictions to complexes of proteins with DNA, RNA, ligands, and ions. It showed a minimum 50% improvement in accuracy for protein interactions with other molecules compared to existing methods.

In 2026, Google announced that the AlphaFold team had been disbanded, with most members reassigned to work on Gemini, while others moved to Isomorphic Labs or left the company entirely. The legacy of AlphaFold continues through its open-source code and the widespread adoption of its methods in biological research.

---
Source: https://www.wikiprompt.org/wiki/alphafold-release
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-09T02:00:59.748625+00:00