AlphaFold 1 (2018) placed first in the overall rankings of the 13thCritical Assessment of Structure Prediction (CASP) in December 2018. It was particularly successful at predicting the most accurate structures for targets rated as most difficult by the competition organizers, where no existingtemplate structures were available fromproteins with partially similar sequences.
AlphaFold 2 (2020) repeated this placement in the CASP14 competition in November 2020.[3] It achieved a level of accuracy much higher than any other entry.[2][4] It scored above 90 on CASP'sglobal distance test (GDT) for approximately two-thirds of the proteins, a test measuring the similarity between a computationally predicted structure and the experimentally determined structure, where 100 represents a complete match.[2][5] The inclusion ofmetagenomic data has improved the quality of the prediction ofMSAs. One of the biggest sources of the training data was the custom-built Big Fantastic Database (BFD) of 65,983,866 protein families, represented as MSAs andhidden Markov models (HMMs), covering 2,204,359,010 protein sequences from reference databases, metagenomes, and metatranscriptomes.[6]
AlphaFold 2's results at CASP14 were described as "astounding"[7] and "transformational".[8] However, some researchers noted that the accuracy was insufficient for a third of its predictions, and that it did not reveal the underlying mechanism or rules ofprotein folding for theprotein folding problem, which remains unsolved.[9][10]
Despite this, the technical achievement was widely recognized. On 15 July 2021, the AlphaFold 2 paper was published inNature as an advance access publication alongsideopen source software and a searchable database of speciesproteomes.[6][11][12] As of February 2025, the paper had been cited nearly 35,000 times.[13]
AlphaFold 3 was announced on 8 May 2024. It can predict the structure ofcomplexes created by proteins withDNA,RNA, variousligands, andions.[14][15] The new prediction method shows a minimum 50% improvement in accuracy for protein interactions with other molecules compared to existing methods. Moreover, for certain key categories of interactions, the prediction accuracy has effectively doubled.[16]
Protein structures can be determined experimentally through techniques such asX-ray crystallography,cryo-electron microscopy andnuclear magnetic resonance (NMR), which are all expensive and time-consuming.[20] Such efforts, using the experimental methods, have identified the structures of about 170,000 proteins over the last 60 years, while there are over 200 million known proteins across all life forms.[5]
Over the years, researchers have applied numerous computational methods topredict the 3D structures of proteins from their amino acid sequences, accuracy of such methods in best possible scenario is close to experimental techniques (NMR) by the use ofhomology modeling based on molecular evolution.CASP, which was launched in 1994 to challenge the scientific community to produce their best protein structure predictions, found thatGDT scores of only about 40 out of 100 can be achieved for the most difficult proteins by 2016.[5] AlphaFold started competing in the 2018 CASP using anartificial intelligence (AI)deep learning technique.[20]
DeepMind is known to have trained the program on over 170,000 proteins from theProtein Data Bank, a public repository of protein sequences and structures. The program uses a form ofattention network, adeep learning technique that focuses on having theAI identify parts of a larger problem, then piece it together to obtain the overall solution.[2] The overall training was conducted on processing power between 100 and 200GPUs.[2]
AlphaFold 1 (2018) was built on work developed by various teams in the 2010s, work that looked at the large databanks of related DNA sequences now available from many different organisms (most without known 3D structures), to try to find changes at differentresidues (peptides) that appeared to be correlated, even though the residues were not consecutive in the main chain. Such correlations suggest that the residues may be close to each other physically, even though not close in the sequence, allowing acontact map to be estimated. Building on recent work prior to 2018, AlphaFold 1 extended this by estimating a probability distribution for the distances between residues, effectively transforming the contact map into a distance map. It also used more advanced learning methods than previously to develop the inference.[21][22]
AlphaFold 2 performance, experiments, and architecture[23]Architectural details of AlphaFold 2[23]
The 2020 version of the program (AlphaFold 2, 2020) is significantly different from the original version that won CASP 13 in 2018, according to the team at DeepMind.[24][25]
AlphaFold 1 used a number of separately trained modules to produce a guide potential, which was then combined with a physics-based energy potential. AlphaFold 2 replaced this with a system of interconnected sub-networks, forming a single, differentiable, end-to-end model based on pattern recognition. This model was trained in an integrated manner.[25][26] After the neural network's prediction converges, a final refinement step applies local physical constraints using energy minimization based on theAMBER force field. This step only slightly adjusts the predicted structure.[27]
A key part of the 2020 system are two modules, believed to be based on atransformer design, which are used to progressively refine avector of information for each relationship (or "edge" in graph-theory terminology) between anamino acid residue of the protein and another amino acid residue (these relationships are represented by the array shown in green); and between each amino acid position and each different sequences in the inputsequence alignment (these relationships are represented by the array shown in red).[26] Internally these refinement transformations contain layers that have the effect of bringing relevant data together and filtering out irrelevant data (the "attention mechanism") for these relationships, in a context-dependent way, learnt from training data. These transformations are iterated, the updated information output by one step becoming the input of the next, with the sharpened residue/residue information feeding into the update of the residue/sequence information, and then the improved residue/sequence information feeding into the update of the residue/residue information.[26] As the iteration progresses, according to one report, the "attention algorithm ... mimics the way a person might assemble a jigsaw puzzle: first connecting pieces in small clumps—in this case clusters of amino acids—and then searching for ways to join the clumps in a larger whole."[5][needs update]
The output of these iterations then informs the final structure prediction module,[26] which also uses transformers,[28] and is itself then iterated. In an example presented by DeepMind, the structure prediction module achieved a correct topology for the target protein on its first iteration, scored as having a GDT_TS of 78, but with a large number (90%) of stereochemical violations – i.e. unphysical bond angles or lengths. With subsequent iterations the number of stereochemical violations fell. By the third iteration the GDT_TS of the prediction was approaching 90, and by the eighth iteration the number of stereochemical violations was approaching zero.[29]
The training data was originally restricted to single peptide chains. However, the October 2021 update, named AlphaFold-Multimer, included protein complexes in its training data. DeepMind stated this update succeeded about 70% of the time at accurately predicting protein-protein interactions.[30]
AlphaFold 3 introduces the "Pairformer," a deep learning architecture inspired by the transformer, which is considered similar to, but simpler than, the Evoformer used in AlphaFold 2.[15][32] The Pairformer module's initial predictions are refined by adiffusion model. This model begins with a cloud of atoms and iteratively refines their positions, guided by the Pairformer's output, to generate a 3D representation of the molecular structure.[14]
The AlphaFold server was created to provide free access to AlphaFold 3 for non-commercial research.[33] As of May 2025, the AlphaFold 3 research paper has been directly cited more than 4000 times.[34]
Results achieved for protein prediction by the best reconstructions in the CASP 2018 competition (small circles) and CASP 2020 competition (large circles), compared with results achieved in previous years. The crimson trend-line shows how a handful of models including AlphaFold 1 achieved a significant step-change in 2018 over the rate of progress that had previously been achieved, particularly in respect of the protein sequences considered the most difficult to predict. (Qualitative improvement had been made in earlier years, but it is only as changes bring structures within 8Å of their experimental positions that they start to affect the CASP GDS-TS measure). The orange trend-line shows that by 2020 online prediction servers had been able to learn from and match this performance, while the best other groups (green curve) had on average been able to make some improvements on it. However, the black trend curve shows the degree to which AlphaFold 2 had surpassed this again in 2020, across the board. The detailed spread of data points indicates the degree of consistency or variation achieved by AlphaFold. Outliers represent the handful of sequences for which it did not make such a successful prediction.
The program was particularly successfully predicting the most accurate structure for targets rated as the most difficult by the competition organisers, where no existingtemplate structures were available from proteins with a partially similar sequence. AlphaFold gave the best prediction for 25 out of 43 protein targets in this class,[36][37][38] achieving a median score of 58.9 on the CASP'sglobal distance test (GDT) score, ahead of 52.5 and 52.4 by the two next best-placed teams,[39] who were also using deep learning to estimate contact distances.[40][41] Overall, across all targets, AlphaFold 1 achieved a GDT score of 68.5.[42]
In January 2020, implementations and illustrative code of AlphaFold 1 was releasedopen-source onGitHub.[43][20] but, as stated in the "Read Me" file on that website: "This code can't be used to predict structure of an arbitrary protein sequence. It can be used to predict structure only on the CASP13 dataset (links below). The feature generation code is tightly coupled to our internal infrastructure as well as external tools, hence we are unable to open-source it." Therefore, in essence, the code deposited is not suitable for general use but only for the CASP13 proteins. The company has not announced plans to make their code publicly available as of 5 March 2021.
In November 2020, DeepMind's new version, AlphaFold 2, won CASP14.[44][45] Overall, AlphaFold 2 made the best prediction for 88 out of the 97 targets.[7]
On the competition's preferredglobal distance test (GDT) measure of accuracy, the program achieved a median score of 92.4 (out of 100), meaning that more than half of its predictions were scored at better than 92.4% for having their atoms in more-or-less the right place,[46][47] a level of accuracy reported to be comparable to experimental techniques likeX-ray crystallography.[24][8][42] In 2018 AlphaFold 1 had only reached this level of accuracy in two of all of its predictions.[7] 88% of predictions in the 2020 competition had a GDT_TS score of more than 80. On the group of targets classed as the most difficult, AlphaFold 2 achieved a median score of 87.[citation needed]
Measured by theroot-mean-square deviation (RMS-D) of the placement of the alpha-carbon atoms of the protein backbone chain, which tends to be dominated by the performance of the worst-fitted outliers, 88% of AlphaFold 2's predictions had an RMS deviation of less than 4Å for the set of overlapped C-alpha atoms.[7] 76% of predictions achieved better than 3 Å, and 46% had a C-alpha atom RMS accuracy better than 2 Å,[7] with a median RMS deviation in its predictions of 2.1 Å for a set of overlapped CA atoms.[7] AlphaFold 2 also achieved an accuracy in modelling surfaceside chains described as "really really extraordinary".
To further validate AlphaFold 2, the conference organizers approached four leading experimental groups working on structures they found particularly challenging and had been unable to determine. In all four cases the three-dimensional models produced by AlphaFold 2 were sufficiently accurate to determine structures of these proteins bymolecular replacement. These included target T1100 (Af1503), a smallmembrane protein studied by experimentalists for ten years.[5]
Of the three structures that AlphaFold 2 had the least success in predicting, two had been obtained byprotein NMR methods, which define protein structure directly in aqueous solution, whereas AlphaFold was mostly trained onprotein structures in crystals. The third exists in nature as amultidomain complex consisting of 52 identical copies of the samedomain, a situation AlphaFold was not programmed to consider. For all targets with a single domain, excluding only one very large protein and the two structures determined by NMR, AlphaFold 2 achieved a GDT_TS score of over 80.
AlphaFold 2 scoring more than 90 inCASP'sglobal distance test (GDT) is a great achievement incomputational biology[5].[8]Nobel Prize winner andstructural biologistVenki Ramakrishnan called the result "a stunning advance on the protein folding problem",[5] adding that "It has occurred decades before many people in the field would have predicted. It will be exciting to see the many ways in which it will fundamentally change biological research."[44]
AlphaFold 2's success received wide media attention.[49][44][50] News pieces appeared in the science press, such asNature,[8]Science,[5]MIT Technology Review,[2] andNew Scientist,[51][52] and the story was covered by national newspapers.[53][54][55][56] A frequent theme was the ability to predict protein structures based on the constituent amino acid sequence, expected to have benefits in the life sciences--accelerating drug discovery and enabling better understanding of diseases.[8][57] Some have noted that even a perfect answer to the proteinprediction problem still leaves questions about the proteinfolding problem (and thusprotein dynamics)—understanding in detail how the folding process actually occurs in nature (and how sometimes they can alsomisfold).[58]
Open access to source code of several AlphaFold versions (excluding AlphaFold 3) has been provided by DeepMind after requests from the scientific community.[61][62][63] The source code of AlphaFold 3[64] was made available for non-commercial use to the scientific community upon request in November 2024.
TheAlphaFold Protein Structure Database, a joint project between AlphaFold andEMBL-EBI, was launched on July 22, 2021. At launch, the database contained AlphaFold-predictedmodels for nearly the completeUniProtproteome of humans and 20model organisms, totaling over 365,000 proteins. The database does not include proteins with fewer than 16 or more than 2700amino acid residues,[65] but for humans they are available in the whole batch file.[66] AlphaFold's initial goal (as of early 2022) was to expand the database to cover most of the UniRef90 set, which contains over 100 million proteins. As of May 15, 2022, the database contained 992,316 predictions.[67]
In July 2021, UniProt-KB andInterPro[68] has been updated to show AlphaFold predictions when available.[69]
On July 28, 2022, the team uploaded to the database the structures of around 200 million proteins from 1 million species, covering nearly every known protein on the planet.[70]
Alphafold-2 was validated for predicting effects of point mutations on structure[73] and free energy[74], with a partial success.
The model relies, to some extent, on co-evolutionary information from similar proteins. Therefore, it may not perform as well on synthetic proteins or proteins with very low homology to those in the training database.[75]
The model's ability to predict multiplenative conformations of proteins is limited.
AlphaFold 3 version can predict structures of protein complexes with a very limited set of selectedcofactors and co- andpost-translational modifications.[76] Between 50% and 70% of the structures of the human proteome are incomplete without covalently-attached glycans.[77][71] AlphaFill, a derived database, adds cofactors to AlphaFold models where appropriate.[78]
In the algorithm, the residues are moved freely, without any restraints. Therefore, during modeling the integrity of the chain is not maintained. As a result, AlphaFold may produce topologically wrong results, like structures with an arbitrary number of knots.[79]
AlphaFold has been used to predict structures of proteins ofSARS-CoV-2, the causative agent ofCOVID-19. The structures of these proteins were pending experimental detection in early 2020.[80][8] Results were reviewed by scientists at theFrancis Crick Institute in the United Kingdom before being released to the broader research community. The team also confirmed accurate prediction against the experimentally determined SARS-CoV-2spike protein that was shared in theProtein Data Bank, an international open-access database, before releasing the computationally determined structures of the under-studied protein molecules.[81] The team acknowledged that although these protein structures might not be the subject of ongoing therapeutical research efforts, they will add to the community's understanding of the SARS-CoV-2 virus.[81] Specifically, AlphaFold 2's prediction of the structure of theORF3a protein was very similar to the structure determined by researchers atUniversity of California, Berkeley usingcryo-electron microscopy. This specific protein is believed to assist the virus in breaking out of the host cell once it replicates. This protein is also believed to play a role in triggering the inflammatory response to the infection.[82]
^"Press release: The Nobel Prize in Chemistry 2024". The Royal Swedish Academy of Sciences. 9 October 2024. Retrieved29 November 2024.The Royal Swedish Academy of Sciences has decided to award the Nobel Prize in Chemistry 2024 with one half to David Baker..."for computational protein design" and the other half jointly to Demis Hassabis... John Jumper..."for protein structure prediction"
^For the GDT_TS measure used, each atom in the prediction scores a quarter of a point if it is within 8 Å (0.80 nm) of the experimental position; half a point if it is within 4 Å, three-quarters of a point if it is within 2 Å, and a whole point if it is within 1 Å.
^To achieve a GDT_TS score of 92.5, mathematically at least 70% of the structure must be accurate to within 1 Å, and at least 85% must be accurate to within 2 Å,
_artificial_intelligence_icon.png){kind=icon}
