# John Jumper

> ~1985– · Computer Scientist, AI Researcher
>
> **Recorded contribution:** AlphaFold 2 lead — solved protein structure prediction; Nobel Prize in Chemistry (2024)

## How to use this dossier

Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory.

## 1. Historical orientation

American chemist and computational biologist John Jumper led the AlphaFold2 research team at DeepMind. The system's 2020 CASP performance and 2021 Nature paper transformed protein-structure prediction by combining evolutionary sequence evidence, attention-based representation, geometric reasoning, and iterative refinement. Jumper and Demis Hassabis shared half of the 2024 Nobel Prize in Chemistry; David Baker received the other half. AlphaFold2 treated protein structure as an inference problem over relationships among residues, integrating evolutionary sequence evidence and geometric constraints. Its Evoformer repeatedly exchanged information between multiple-sequence and pair representations before a structure module produced three-dimensional coordinates and confidence estimates.

## 2. The problem inherited

Determining protein structures experimentally is slow and expensive, while sequence-to-structure prediction had remained inaccurate for many proteins despite decades of research.

## 3. The central contribution

Jumper led AlphaFold2's model design and research execution, producing highly accurate structure predictions at unprecedented breadth.

## 4. Reconstruct the mechanism

1. Collect a target amino-acid sequence, related sequences, and available structural templates.
2. Use an Evoformer to exchange information between sequence alignments and pairwise residue representations.
3. Predict three-dimensional residue geometry with a structure module and recycle predictions through the network.
4. Report coordinates with confidence estimates that help users distinguish stronger from weaker regions.

## 5. What changed downstream

- AlphaFold predictions became widely used hypotheses for structural biology and drug and protein research.
- The public database dramatically broadened access to predicted structures.
- The system greatly expanded practical access to predicted structures, changing hypothesis generation in structural biology and enabling databases covering proteins that lacked experimentally determined models.

## 6. Attribution, limits, and uncertainty

- AlphaFold2 has many authors and depends on experimental databases, sequence repositories, compute, and decades of structural biology.
- A static prediction does not fully capture dynamics, ligands, complexes, disorder, cellular context, function, or experimental validation.
- A prediction is not a complete account of dynamics, binding, disorder, environment, or biological function; confidence scores require interpretation and experimental structure determination remains essential.

## 7. Reconstruction lab

Choose one public AlphaFold prediction, color it by confidence, compare it with an experimental structure if available, and list every scientific claim the prediction alone cannot support. Select one high-confidence and one low-confidence region, compare both with experimental evidence, and write different claims that each confidence level actually supports.

## 8. Evidence trail

- [John Jumper — facts](https://www.nobelprize.org/prizes/chemistry/2024/jumper/facts/) — Nobel Prize
- [Highly accurate protein structure prediction with AlphaFold](https://www.nature.com/articles/s41586-021-03819-2) — Nature
- [AlphaFold Protein Structure Database](https://alphafold.ebi.ac.uk/) — EMBL-EBI and Google DeepMind

---

*Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*