Today's Issue

Main Topic: What proteins are, why their shape matters so much, what DeepMind built, how it works, and what it means for medicine and drug discovery

Subtitles:

• Proteins: the machines that run your body

• The 50-year problem AlphaFold solved

• How AlphaFold actually works

• What it means for medicine and drug discovery

• What comes next: AlphaFold 3 and the era of digital biology

Abstract: Proteins are the fundamental molecular machines of all living organisms, made from chains of amino acids (small organic molecules, 20 types in total) that fold into precise three-dimensional shapes. The shape of a protein determines its function: a misfolded protein cannot do its job and often causes disease. Determining protein structure experimentally using X-ray crystallography or cryo-electron microscopy (a technique that fires electrons at frozen protein samples to image them) takes years of work and hundreds of thousands of dollars per protein. The "protein folding problem," the challenge of predicting a protein's 3D shape from its amino acid sequence alone, was considered one of biology's central unsolved challenges for five decades. AlphaFold 2, developed by DeepMind (London, acquired by Alphabet in 2014), won the CASP14 protein structure prediction competition in 2020 with predictions achieving median error below 1 Angstrom (0.1 nanometers), three times more accurate than the next best system, and comparable to experimental methods. In 2022, DeepMind released predictions for over 200 million protein structures, covering virtually all catalogued proteins known to science, freely available through the AlphaFold Protein Structure Database. The database has been used by over 3 million researchers in more than 190 countries. Over 30% of AlphaFold-related research focuses on disease understanding. Research linked to AlphaFold 2 is twice as likely to be cited in clinical articles than typical structural biology work. AlphaFold 3 (2024), developed with Isomorphic Labs, extends predictions beyond proteins to all biomolecules including DNA, RNA, and ligands (the small molecules that drugs are based on), enabling modeling of how drugs interact with their protein targets. The 2024 Nobel Prize in Chemistry was awarded to DeepMind's Demis Hassabis and John Jumper for AlphaFold, the first Nobel Prize ever awarded for an AI-enabled scientific breakthrough.

The Tech newsletter for Engineers who want to stay ahead

Tech moves fast, but you're still playing catch-up?

That's exactly why 200K+ engineers working at Google, Meta, and Apple read The Code twice a week.

Here's what you get:

• Curated tech news that shapes your career - Filtered from thousands of sources so you know what's coming 6 months early.

• Practical resources you can use immediately - Real tutorials and tools that solve actual engineering problems.

• Research papers and insights decoded - We break down complex tech so you understand what matters.

All delivered twice a week in just 2 short emails.

Join 200K+ engineers

1. Proteins: The Machines That Run Your Body 🧬⚙️

Your body contains roughly 100,000 different types of proteins.

They do almost everything.

Enzymes are proteins that speed up chemical reactions.

Antibodies are proteins that identify and neutralize pathogens.

Hemoglobin is a protein that carries oxygen through your blood. Receptors on your cell surfaces are proteins.

Hormones like insulin are proteins. The structural scaffolding of your cells, collagen in your skin, actin in your muscles, all proteins.

Each protein is built from a chain of amino acids (small organic molecules, 20 types total), assembled according to instructions from your DNA.

But here is the critical part: it is not just the sequence that matters. It is the shape.

Once a protein chain is assembled, it folds spontaneously into a precise three-dimensional structure. That specific shape determines whether the protein can do its job, whether it can bind to the right molecule, fit into the right receptor, catalyze the right reaction.

A misfolded protein is like a key cut incorrectly. It cannot open the lock.

Protein misfolding is directly implicated in Alzheimer's disease (where tau and amyloid proteins misfold into toxic tangles), Parkinson's disease, Type 2 diabetes, cystic fibrosis, and many cancers.

2. The 50-Year Problem AlphaFold Solved 🔬⏳

Knowing the amino acid sequence of a protein is relatively easy.

Working out the 3D shape it folds into? That was the problem.

The traditional method, X-ray crystallography (a technique where protein crystals are bombarded with X-rays and the diffraction pattern reveals the atomic structure), requires growing pure protein crystals (crystal structure is KEY), which is extremely difficult for many proteins, then months to years of analysis.

Another method, cryo-electron microscopy (which fires electrons at flash-frozen protein samples to image their shape), is faster but still expensive and technically demanding.

The number of known protein sequences has always massively outpaced the number of experimentally determined structures. There are hundreds of millions of known protein sequences and, before AlphaFold, only about 170,000 known structures. A huge gap.

In 2020, AlphaFold entered the CASP14 competition, the biennial international contest for protein structure prediction, described as the Olympics of structural biology.

Its predictions achieved a median error below 1 Angstrom (0.1 nanometers, roughly the size of a single atom) and were three times more accurate than the next best system, described by competition organizers as a solution to the 50-year-old protein folding problem.

3. How AlphaFold Actually Works 🤖🔍

AlphaFold does not brute-force its way through every possible folding configuration.

It uses a deep learning architecture (a type of AI modeled loosely on how the brain processes information) called the Evoformer.

The key insight: evolution has already done much of the work.

If two species both have a protein that performs the same function, and both proteins have slightly different amino acid sequences, the differences that evolution has preserved (and the positions that have changed together) reveal which amino acids are physically close to each other in the folded structure.

AlphaFold takes a protein's amino acid sequence, searches databases of millions of evolutionarily related sequences from other organisms, and uses the patterns of co-evolution to infer the spatial relationships between amino acids.

It then iteratively refines a 3D structure prediction, checking it against what it knows about how proteins physically behave, until it converges on a shape.

A structure that once took years to determine experimentally now takes minutes on AlphaFold.

In 2022, DeepMind released the AlphaFold Protein Structure Database in partnership with EMBL-EBI (the European Bioinformatics Institute), containing predictions for over 200 million proteins covering virtually all catalogued proteins known to science. Freely available to any researcher anywhere in the world.

4. What It Means for Medicine and Drug Discovery 💊🔬

This is where it becomes genuinely transformative.

Drug discovery is, at its core, a shape-matching problem. Most drugs work by binding to a specific protein (the "target") and either blocking it, activating it, or changing its behavior.

To design a drug that fits a target precisely, you need to know the exact 3D shape of that target.

Before AlphaFold, a significant proportion of potential drug targets were what researchers called "undruggable," not because drugs against them were impossible, but because no one knew their shape well enough to design one.

Some specific examples already emerging.

Research linked to AlphaFold is twice as likely to be cited in clinical articles compared to typical structural biology work. Over 30% of all AlphaFold-related research is directly focused on disease.

Three million researchers across 190 countries are using the database, including over 1 million in low and middle-income countries who previously had no access to the expensive equipment needed for experimental structure determination.

5. What Comes Next: AlphaFold 3 and the Era of Digital Biology 🚀🌐

AlphaFold 2 solved proteins. AlphaFold 3, released in 2024, went further.

AlphaFold 3 predicts the structure and interactions of virtually all life's molecules: proteins, DNA, RNA, ligands (the small molecules that drugs are made of), ions, and the chemical modifications that regulate how cells function.

This matters enormously for drug discovery. A drug molecule does not exist in isolation. It needs to bind to a specific protein, at a specific location, in a specific orientation. AlphaFold 3 can model that entire interaction, protein and drug together, with unprecedented accuracy.

Isomorphic Labs, an AI drug discovery company spun out of DeepMind in 2021 specifically because of AlphaFold's potential, is using AlphaFold 3 as the engine of an entirely new approach to drug design, starting from the molecular interaction and working backward to the drug.

In October 2024, the Nobel Prize in Chemistry was awarded to DeepMind's Demis Hassabis (CEO) and John Jumper (lead researcher on AlphaFold), alongside protein design pioneer David Baker from the University of Washington. It was the first Nobel Prize ever awarded for an AI-enabled scientific breakthrough.

DeepMind's own framing of what comes next: biology is their first frontier, and AlphaFold is the template for how AI can accelerate all of science, from chemistry to climate to energy.

The next AlphaFold-scale breakthrough, in some other domain, is likely already being built somewhere.

Takeaways

• Proteins are the molecular machines of life, and their function is entirely determined by their 3D shape: misfolded proteins directly cause Alzheimer's, Parkinson's, Type 2 diabetes, and many cancers; determining those shapes experimentally (via X-ray crystallography or cryo-electron microscopy) took years and hundreds of thousands of dollars per protein, which is why the 50-year unsolved protein folding problem, predicting a protein's shape from its amino acid sequence alone, was considered one of the central challenges in all of biology.

• DeepMind's AlphaFold 2 won the CASP14 international prediction competition in 2020 with accuracy three times better than any previous system, achieving errors below 1 Angstrom, comparable to experimental methods; by 2022 the freely available AlphaFold database contained predictions for over 200 million proteins covering virtually all catalogued proteins on earth, used by 3 million researchers in 190 countries, with over 30% of associated research directly targeting disease and work linked to AlphaFold twice as likely to be cited in clinical articles than typical structural biology.

• AlphaFold 3 (2024) extended the system beyond proteins to all biomolecules including DNA, RNA, and drug molecules, enabling modeling of how drugs interact with their targets and opening the door to designing medicines for previously "undruggable" proteins; the 2024 Nobel Prize in Chemistry awarded to DeepMind's Demis Hassabis and John Jumper was the first Nobel Prize ever given for an AI-enabled scientific breakthrough, and Isomorphic Labs, the drug discovery spinout built on AlphaFold's foundation, has explicitly stated its goal of one day solving all diseases.

Go from AI overwhelmed to AI savvy professional

AI will eliminate 300 million jobs in the next 5 years.

Yours doesn't have to be one of them.

Here's how to future-proof your career:

• Join the Superhuman AI newsletter - read by 1M+ professionals

• Learn AI skills in 3 mins a day

• Become the AI expert on your team

Start learning AI now