Pull to refresh
Logo
Daily Brief
Following Why Ranks Sign Up
AI closes in on biology's last structural puzzles

AI closes in on biology's last structural puzzles

New Capabilities

From 50-year grand challenge to minutes-long computation, protein prediction enters its commercial era

March 4th, 2026: AlphaFold 4 predicts protein complexes in minutes

Overview

Determining the three-dimensional shape of a protein used to take a graduate student an entire career. Google DeepMind released AlphaFold 4 in March 2026, extending progress from the original AlphaFold's 2018 win, with the system predicting multi-protein structures in minutes instead of hours or days.

Weeks earlier, Isomorphic Labs disclosed IsoDDE, a proprietary model that more than doubles AlphaFold 3's accuracy on drug-design benchmarks. A team at the National University of Singapore published D-I-TASSER, which outperforms both AlphaFold 2 and AlphaFold 3 on difficult multi-domain proteins.

The practical stakes are enormous: drug development averages more than a decade and roughly two billion dollars per approved medicine, with structure determination representing a critical bottleneck in identifying how molecules bind to disease targets. AI systems that predict these interactions in seconds rather than months could compress discovery timelines dramatically. This sparks debate: should the tools remain accessible to all researchers, or become advantages for well-funded companies?

Questions about this story

No questions yet — be the first to ask.

Play on this story Voices Debate Predict

Key Indicators

214M+
Protein structures predicted
AlphaFold's public database now covers virtually every catalogued protein known to science
2x
IsoDDE accuracy gain over AlphaFold 3
Isomorphic Labs' proprietary model more than doubles AlphaFold 3's accuracy on protein-ligand benchmarks
2M+
Researchers using AlphaFold
Scientists across 190 countries have used the AlphaFold database for published research
$3.3B
AI drug discovery funding in 2024
Venture capital invested in artificial intelligence-driven drug discovery startups in a single year

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

Log in Sign Up

Play

Exploring all sides of a story is often best achieved with Play.

Log in to play. Track your picks, climb the leaderboards. Log in Sign Up
Predict 4 ways this could play out. Contrarian picks score more — points lock when the scenario resolves. Log in to play
Timeline Five events from this story — drag them oldest to newest. Log in to play
Connections Sixteen names from the news. Find the four hidden groups of four. Log in to play

People Involved

Demis Hassabis
Demis Hassabis
2024 Nobel laureate in Chemistry; leads both DeepMind's research and its drug-discovery spinoff
 John Jumper
John Jumper
2024 Nobel laureate in Chemistry; leads AlphaFold development
 David Baker
David Baker
2024 Nobel laureate in Chemistry; leads open-source protein design tools
 Mohammed AlQuraishi
Mohammed AlQuraishi
Developing open-source alternatives to AlphaFold
 Zhang Yang
Zhang Yang
Lead developer of D-I-TASSER protein prediction pipeline

Organizations Involved

Google DeepMind
Google DeepMind
AI research division
Developer of AlphaFold series; released AlphaFold 4 in March 2026

Alphabet's artificial intelligence research laboratory, responsible for developing the AlphaFold series of protein structure prediction models.
Isomorphic Labs
Isomorphic Labs
AI Drug Discovery Company
Released proprietary IsoDDE model in February 2026; partnerships with major pharmaceutical companies

An Alphabet-owned drug discovery company that commercializes protein structure prediction technology developed by Google DeepMind.
European Molecular Biology Laboratory - European Bioinformatics Institute (EMBL-EBI)
European Molecular Biology Laboratory - European Bioinformatics Institute (EMBL-EBI)
Research Institution
Hosts the AlphaFold Protein Structure Database

A European intergovernmental research organization that partners with Google DeepMind to host and distribute the AlphaFold database of over 214 million predicted protein structures.

Timeline

December 2018 March 2026

12 events Latest: March 4th, 2026 · 4 months ago Showing 8 of 12
Tap a bar to jump to that date

  1. AlphaFold 4 predicts protein complexes in minutes

    Latest Breakthrough

    Google DeepMind released AlphaFold 4, which predicts the structure of multi-protein complexes in minutes rather than the hours or days required by its predecessor, using an improved neural network architecture and expanded training data.

  2. D-I-TASSER outperforms AlphaFold on complex proteins

    Research

    A team at the National University of Singapore published D-I-TASSER, a hybrid deep-learning and physics-based tool that predicts complex multi-domain protein structures about 13 percent more accurately than AlphaFold 3.

  3. Isomorphic Labs unveils IsoDDE, a proprietary 'AlphaFold 4'

    Breakthrough

    Isomorphic Labs released a technical report describing IsoDDE, a proprietary drug design engine that more than doubles AlphaFold 3's accuracy on protein-ligand benchmarks. Unlike previous AlphaFold versions, IsoDDE is not publicly available.

  4. Boltz-1 matches AlphaFold 3 performance as fully open model

    Release

    An MIT lab released Boltz-1 under an MIT license with full training code and model weights, matching AlphaFold 3 and Chai-1 across key protein prediction benchmarks.

  5. Nobel Prize in Chemistry awarded for protein prediction

    Recognition

    Demis Hassabis and John Jumper shared half the Nobel Prize in Chemistry for AlphaFold; David Baker received the other half for computational protein design.

  6. Chai-1 released as open-source AlphaFold competitor

    Release

    Chai Discovery released Chai-1 under an Apache 2.0 license, providing an open-source alternative to AlphaFold 3 for academic and commercial use.

  7. AlphaFold 3 predicts multi-molecule interactions

    Breakthrough

    Google DeepMind and Isomorphic Labs released AlphaFold 3, capable of predicting how proteins interact with DNA, RNA, small molecules, and ions, with at least 50 percent better accuracy than existing methods.

  8. AlphaFold database expands to 200 million structures

    Release

    DeepMind and EMBL-EBI uploaded predicted structures for nearly every catalogued protein known to science, covering approximately one million species.

  9. Isomorphic Labs founded

    Corporate

    Demis Hassabis launched Isomorphic Labs as an Alphabet-owned company to apply AlphaFold technology to commercial drug discovery.

  10. AlphaFold database launches with 350,000 structures

    Release

    Google DeepMind and EMBL-EBI launched the AlphaFold Protein Structure Database, initially covering the entire human proteome and 20 model organisms.

  11. AlphaFold 2 solves the protein folding problem

    Breakthrough

    AlphaFold 2 achieved atomic-level accuracy at CASP14, predicting protein structures with a median error of less than one angstrom. Scientists declared the 50-year-old protein folding problem effectively solved.

  12. AlphaFold 1 wins CASP13

    Breakthrough

    DeepMind's AlphaFold placed first in the 13th Critical Assessment of Structure Prediction competition, outperforming all other computational methods for predicting protein structures from amino acid sequences.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

1990-2003

Human Genome Project (1990-2003)

A 13-year, $2.7 billion international effort sequenced the roughly 3 billion base pairs of human DNA. The project competed with Craig Venter's privately funded Celera Genomics, which used a faster shotgun sequencing method and announced a draft genome simultaneously with the public effort in 2000.

Then

Both the public consortium and Celera published draft genomes in February 2001. The public data was made freely available, while Celera initially restricted access through a subscription model.

Now

The open-access model won. Free genome data enabled thousands of studies that Celera's proprietary model could not have supported. The project catalyzed the genomics industry, drove DNA sequencing costs from $100 million per genome in 2001 to under $200 by 2024, and established a precedent for open science in biology.

Why this matters now

The tension between Isomorphic Labs' proprietary IsoDDE and the open AlphaFold database mirrors the Celera-versus-public-consortium dynamic. The genome project showed that open access to foundational biological data produces more aggregate innovation, a lesson that shapes the current debate over whether AI drug design tools should be public goods.

2012-present

CRISPR gene editing revolution (2012-present)

Jennifer Doudna and Emmanuelle Charpentier published their discovery that a bacterial immune system component, CRISPR-Cas9, could be reprogrammed to cut any DNA sequence with precision. The technology reduced the cost and time of genetic editing from months and thousands of dollars to days and a few hundred dollars.

Then

The discovery triggered a patent battle between the Broad Institute and UC Berkeley, a 2020 Nobel Prize in Chemistry for Doudna and Charpentier, and rapid commercialization by companies like Editas Medicine and CRISPR Therapeutics.

Now

In December 2023, the United States Food and Drug Administration approved the first CRISPR-based therapy, Casgevy, for sickle cell disease. The technology moved from laboratory discovery to approved medicine in roughly 11 years, a timeline that AI-driven drug design aims to compress further.

Why this matters now

Like CRISPR, AI protein prediction represents a platform technology that dramatically lowers the barrier to biological manipulation. Both compress timelines from years to days and raise questions about equitable access. The CRISPR patent wars offer a preview of the intellectual property disputes likely to emerge as proprietary AI models become central to drug development.

1958

X-ray crystallography and the first protein structure (1958)

John Kendrew and Max Perutz at Cambridge used X-ray crystallography to determine the three-dimensional structure of myoglobin, a protein that stores oxygen in muscle tissue. The work required growing protein crystals, bombarding them with X-rays, and painstakingly interpreting the resulting diffraction patterns. It took Kendrew roughly two decades of effort.

Then

Kendrew and Perutz received the 1962 Nobel Prize in Chemistry. X-ray crystallography became the dominant method for determining protein structures for the next six decades.

Now

By 2020, researchers had determined approximately 170,000 protein structures through experimental methods. That number represented a tiny fraction of the estimated 200 million known proteins, illustrating the bottleneck that AlphaFold ultimately broke.

Why this matters now

The contrast in scale is the point. What took Kendrew 20 years for a single protein, AlphaFold 4 now accomplishes for multi-protein complexes in minutes. The roughly 170,000 experimentally determined structures over six decades compare to 214 million computationally predicted structures released in a single database expansion.

Sources

(12)
+6
abit.ee Nature Isomorphic Labs Phys.org Nobel Prize Google Blog Science EMBL-EBI Google DeepMind Scientific American ScienceDaily Nature Biotechnology