AlphaFold

basicwidespread

AlphaFold is Google DeepMind's artificial intelligence system that predicts the 3D structure of a protein based on the sequence of its building blocks.

In plain terms

Your body performs almost every process with proteins: they digest, transport oxygen, and defend against pathogens. A protein is a long chain that folds into a ball-like shape, and its shape determines what it can do. It is like a key: the material does not matter, but the shape does, because that fits into the lock.

The sequence of building blocks forming the chain has long been easy to read, but determining the shape required expensive laboratory experiments lasting months or years. AlphaFold replaces this step: it estimates the spatial structure from the sequence in minutes or hours, often with accuracy rivaling laboratory results.

It is not a chatbot, nor a general-purpose AI. It was built for a single scientific task and delivered a breakthrough there. This is why it is often cited as an example of how artificial intelligence can accelerate research, not just generate text and images.

How it works

It receives the protein's 'recipe', meaning the amino acid sequence as a string of letters.
It searches databases for related proteins from other organisms and examines which building blocks changed together during evolution, as these are likely close to each other in space.
A neural network trained on experimentally measured protein structures calculates the relative positions of building blocks from this and refines the spatial model over multiple iterations.
At the end, it assigns a reliability score to each part, so the researcher sees where to trust the prediction and where uncertainty remains.
AlphaFold 3, released in 2024, now also estimates how proteins bind to other molecules, such as DNA or drug active ingredients.

An everyday example

You read a news report that researchers are working on a new antibiotic or plastic-degrading enzyme, and the article mentions that the target protein structure was determined using artificial intelligence. Behind this, there is a high chance it is AlphaFold or the freely searchable AlphaFold database, from which the researcher downloaded the structure with a few clicks.

Who is affected

patient, doctorEarly-stage drug and vaccine research can be faster if the shapes of proteins involved in the disease are already known. → Health

parent, student, teacherIn biology and chemistry education, as well as university research, it is a free tool and visual material available to anyone. → Education & Family

industrial worker, engineerIt has become a daily tool in biotech, food, and chemical industry developments that design enzymes. → Industry

bank customer, investorBiotech companies are often evaluated with reference to it, so it is worth knowing what it solves and what it does not. → Money

A common misconception

MythAlphaFold discovers drugs and has solved biology.

In realityIt only predicts protein shapes, which is just one early step in research. Finding the drug, conducting lab tests, and running clinical trials still take years.

Limits and risks

The result is a calculated prediction, not a measurement: it must be verified with an experiment before important decisions.
It typically provides a single, rigid shape, whereas proteins move and change form; it performs poorly in disordered, flexible regions.
It can hardly say how a mutation changes the structure, nor what the protein actually does in the cell.
Commercial use of AlphaFold 3 is limited, so not everyone has equal access to the latest version.

Similar, but not the same

AlphaGo – AlphaGo is the Go-playing program from the same company, while AlphaFold solves a scientific task rather than playing a game.
Large language model – A language model writes text and answers questions, while AlphaFold exclusively calculates the spatial structure of molecules.
Protein design – AlphaFold predicts a shape for a given building-block sequence, while protein design tools work in reverse: they create new, non-natural proteins for a desired shape or function.

Related concepts

Broader

Artificial intelligence, Deep learning, AI in science

Narrower

AlphaFold 3, AlphaFold protein structure database

On the edge

Google DeepMind, AlphaGo, Neural network, Transzformer, Drug research with artificial intelligence

A brief history

DeepMind's first AlphaFold system appeared in 2018 at the CASP international protein structure prediction competition, and the breakthrough came in 2020 with AlphaFold 2, whose description and code were released in 2021. AlphaFold 3 was released in 2024, and in the same year, Demis Hassabis and John Jumper received half of the Nobel Prize in Chemistry for protein structure prediction; the other half went to David Baker for computer-aided protein design.

Where things stand

In late 2026, it is a basic tool in structural biology: the freely accessible database contains predicted structures of more than 200 million proteins, and millions of researchers worldwide use it. In drug development, it accelerates the early research phase but does not replace clinical trials.

Status: 2026 November

Also known as

AlphaFold · AlphaFold 2 · AlphaFold2 · AlphaFold 3 · AF2 · AF3 · AlphaFold DB · AlphaFold Server

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.