How a single AI system solved a 50-year biology challenge, and why 2026 is the year AlphaFold moves from prediction to real-world medicine.
Public guide12 minute readReviewed August 21, 2026
What is AlphaFold?
AlphaFold is an artificial intelligence system developed by DeepMind (a Google company) that predicts the three-dimensional shape of proteins from their amino acid sequence. Proteins are the molecular machines that do almost everything inside living cells — building structures, catalysing reactions, carrying signals, and defending the body. Knowing a protein's shape is often the key to understanding what it does and how to design drugs that interact with it.
Before AlphaFold, predicting protein structure from sequence alone was one of the biggest unsolved problems in biology. The process took years, required expensive lab equipment like X-ray crystallography or cryo-electron microscopy, and many structures simply could not be determined. AlphaFold changed that timeline from years to seconds.
How it works
AlphaFold does not simply look up known structures. It uses deep learning to infer the physical and evolutionary relationships between amino acids. The system was trained on the Protein Data Bank — a public database of experimentally determined protein structures — and learned the patterns that connect amino acid sequences to the shapes they fold into.
Evolutionary information
AlphaFold searches through databases of related protein sequences across species. Amino acids that mutate together often sit close together in the 3D structure. This co-evolution signal is one of the strongest clues AlphaFold uses.
Attention networks
The system uses a form of neural network called attention, similar to the architectures behind large language models. Attention lets the system weigh which parts of a protein sequence matter most when predicting the position of each atom.
Structure refinement
AlphaFold does not produce a single guess. It generates multiple candidate structures and uses an internal scoring system to select the most physically plausible arrangement. The output includes a confidence score for each region of the protein.
The Nobel Prize
In October 2024, Demis Hassabis (co-founder of DeepMind) and John Jumper (lead architect of AlphaFold) were awarded the Nobel Prize in Chemistry for their work on protein structure prediction. The Nobel Committee recognised that AlphaFold had "solved a major Grand Challenge in biology" and had fundamentally changed how researchers approach protein science.
The prize was notable not just for the achievement itself, but for what it signalled: the scientific establishment had formally recognised AI as a tool capable of delivering fundamental scientific breakthroughs, not just incremental improvements.
What changed in 2026
The AlphaFold project has continued expanding rapidly. In 2025 and 2026, DeepMind released AlphaGenome, which predicts how protein binding interacts with DNA across the entire human genome, and AlphaFold 3, which can predict the structure of proteins bound to DNA, RNA, and small molecules — a critical step toward understanding drug-target interactions.
Meanwhile, Isomorphic Labs (the biotech company spun out from DeepMind) has been using AlphaFold technology to design entirely new drugs from scratch, rather than simply discovering existing ones. Several AlphaFold-designed molecules have entered preclinical development, and the first wave of results is expected by 2027.
Open-source alternatives like OpenFold have also matured significantly, allowing researchers without access to DeepMind's infrastructure to run AlphaFold-quality predictions on their own hardware.
Beyond proteins
AlphaFold's success has sparked a broader wave of AI-driven structural biology. New systems are being developed to predict how multiple proteins interact with each other, how proteins bind to membranes, and how protein shapes change when they interact with drugs or other molecules.
One promising area is protein pairing — predicting how two or more proteins come together to form functional complexes. This matters because most biological processes involve multiple proteins working together, not single proteins in isolation.
Another frontier is protein design — using AI to create entirely new proteins that do not exist in nature. Early results include proteins that bind to disease-causing molecules, degrade harmful substances, or act as new types of vaccine candidates.
How researchers use it
Today, researchers around the world use AlphaFold in three main ways:
Structure lookup — searching the AlphaFold Protein Structure Database (now covering over 200 million predicted structures) to find the shape of any known protein.
Hypothesis generation — using predicted structures to understand why a mutation causes disease, or to identify potential drug targets.
Drug discovery — combining AlphaFold predictions with molecular docking simulations to test how candidate drugs might bind to their targets.
The database is free and open to anyone. You do not need special equipment, training, or permission to explore it.
Go deeper
This guide covers the basics of AlphaFold and why it matters. If you are a chemistry PhD, a biology researcher, or simply someone who wants to understand the frontier of AI in science, the members workspace has five advanced guides that build on this foundation:
Learn AI