Five years after its debut, Google’s DeepMind’s AlphaFold is reshaping biological research, recently earning its developers the Nobel Prize in Chemistry. The artificial intelligence, initially known for mastering games like Go, has instead solved a 50-year grand challenge of science: accurately predicting protein structures-essential building blocks for understanding and combating disease. Now, DeepMind is expanding the AI’s capabilities, moving beyond structure prediction toward simulating complex biological systems and assisting scientists in groundbreaking discoveries.
Five years after its launch, Google’s DeepMind’s AlphaFold is being hailed as a landmark achievement in the history of modern science, and its developers recently received the Nobel Prize in Chemistry for its profound impact on understanding protein structure and accelerating biological discoveries. Understanding proteins is fundamental to understanding and treating disease, making this a significant step forward for medical research.
Prior to AlphaFold’s emergence in late 2020, DeepMind was primarily known for developing artificial intelligence that defeated world champions in electronic games. The company then turned its attention to a more complex and impactful challenge: solving the problem of protein folding, considered one of the most difficult problems in modern science. The result was AlphaFold 2, capable of predicting the three-dimensional structure of proteins with atomic accuracy.
A Database Covering All Proteins
Table of Contents
DeepMind’s work culminated in the creation of a database containing more than 200 million predicted protein structures, representing virtually all known proteins. Currently, approximately 3.5 million researchers in over 190 countries are utilizing this resource. The scientific paper published in the journal Nature in 2021 has garnered over 40,000 citations to date, demonstrating its widespread influence.
Last year, DeepMind unveiled AlphaFold 3, expanding the AI’s scope to include DNA, RNA, and drug design. This ambitious step towards modeling complex biological interactions signals a more comprehensive future for artificial intelligence in the life sciences.
From Games to the Roots of Science
In an interview with WIRED, Bushra Kohli, DeepMind’s Vice President of Research and Engineering Lead for “AI for Science,” explained that the transition from challenges like games to fundamental scientific problems has been part of the company’s vision since its inception.
“Games were a testing ground for developing technologies, but the ultimate goal was always to employ AI to accelerate scientific discovery,” Kohli said, noting that the protein folding problem falls within what he described as “root problems”—issues whose solutions unlock entire fields of research in medicine and biology. This breakthrough has the potential to dramatically accelerate the pace of scientific advancement.
Addressing concerns about “hallucinations” in AI models, particularly with AlphaFold 3’s reliance on diffusion models, Kohli affirmed that the core philosophy remains unchanged, emphasizing a combination of creative generation and rigorous verification.
He added that the system includes confidence indicators that show when predictions are less reliable, especially in inherently irregular proteins. According to Kohli, the scientific confidence in AlphaFold stems from its repeated testing in laboratories and its practical success.
An Intelligent “Scientist Assistant”
DeepMind is now launching an “AI-Powered Scientist Assistant,” built on the Gemini 2.0 model, capable of generating and discussing scientific hypotheses through multiple intelligent agents based on AI models.
Kohli believes this approach doesn’t replace scientists but redistributes their roles, with AI accelerating problem-solving while humans remain responsible for defining which problems are worth solving.
He cited an example of researchers at Imperial College London using the system to study how some viruses hijack bacteria, opening new avenues for combating antibiotic resistance.
Toward Simulating the Human Cell
Looking ahead, Kohli stated that the biggest goal in the coming years is to understand the cell as an integrated unit, ultimately reaching a precise simulation of the entire human cell. He considers deciphering the genome and understanding the mechanisms of reading the genetic code within the cell nucleus to be the first step on this long road.
Kohli concluded that the ability to simulate cells could revolutionize medicine, enabling computer-based drug testing before manufacturing, understanding diseases at a fundamental level, and designing personalized treatments for each patient, directly linking computational predictions to clinical applications that improve patient lives.
