As a physicist who has spent most of my career translating laws of the Universe into lines of code, I’ve seen many technologies promise “revolutions.” But few have hit the scientific community with the force of AlphaFold2.

For fifty years, the “Protein Folding Problem” was the ultimate grand challenge in biology. We knew the sequence of amino acids (the 1D “instruction manual”), but we couldn’t reliably predict the 3D shape they would take. In physics terms, this is an energy minimization problem of astronomical complexity.

The Physics of the Fold

Proteins are the molecular machines of life. Their function is dictated entirely by their geometry. If a protein is a key, its 3D shape is the teeth that must fit a specific lock. Historically, finding this shape required years of grueling lab work—X-ray crystallography or cryo-electron microscopy.

AlphaFold2 changed the game by treating protein folding as a spatial reasoning problem. Instead of just crunching numbers, it uses a transformer-based architecture—the same logic behind LLMs—to attend to the relationships between distant amino acids. It predicts how they will interact across space, effectively learning the “grammar” of molecular physics.

Why Now Was the Turning Point

While the system debuted a few years ago, 2024 was the year the world fully acknowledged its magnitude.

• The Nobel Prize: In October 2024, the Nobel Prize in Chemistry was awarded to Demis Hassabis and John Jumper of Google DeepMind (sharing it with David Baker). This was a historic moment—the first time an AI-driven discovery received the highest honor in science.
• The 200 Million Map: By 2024, the AlphaFold Protein Structure Database (in partnership with EMBL-EBI) had mapped nearly every protein known to science. We went from knowing the shapes of a few hundred thousand proteins to over 200 million.

The Shift to AlphaFold3

As we head into 2025, the focus has shifted from just “folding” to “interacting.”

• Beyond Proteins: While AlphaFold2 was about the protein itself, the new AlphaFold 3 can predict how proteins interact with DNA, RNA, and ligands (the small molecules that become drugs).
• Drug Discovery: This is where my world of computing meets the pharmacy. We can now simulate how a new drug candidate will bind to a target protein with unprecedented accuracy, potentially cutting years off the development cycle for everything from cancer treatments to plastic-eating enzymes.

As my partner Angelia always says: “technology is the ladder, not the destination.” AlphaFold2 is not and should not “doing the science” for us; it is providing the map so we can finally start the journey, it has turned a structural biology bottleneck into a computational data stream.