How machines learned molecular shape

2020AlphaFold 2 stuns CASP14

At the 14th Critical Assessment of protein Structure Prediction, DeepMind's AlphaFold 2 — a team led by John Jumper under CEO Demis Hassabis in London — predicts protein structures at near-experimental accuracy, effectively settling a problem opened in the 1960s: how a chain of amino acids finds its fold. Structural biology changes overnight.

Jumper J. et al., Nature 596, 583 (2021) — abstract

2021The Baker lab opens a second front

David Baker's group at the University of Washington — for decades the home of Rosetta — publishes RoseTTAFold (Minkyung Baek and colleagues, Science), bringing deep-learning structure prediction to academic labs worldwide, and releases methods that turn prediction into design: proteins built from scratch on a computer.

2023RFdiffusion designs from nothing

Joseph Watson, David Juergens, Nathaniel Bennett and colleagues in the Baker lab publish RFdiffusion in Nature: a diffusion model — the same model family behind image generators — that dreams up functional proteins de novo: binders, enzymes' scaffolds, symmetric assemblies. Peptide and protein design becomes generative.

Watson J.L. et al., Nature 620, 1089 (2023) — abstract

2024Peptide binders, AlphaFold 3 — and Stockholm

Susana Vázquez Torres, Philip Leung and colleagues (Baker lab) design high-affinity binders of bioactive helical peptides de novo (Nature) — AI aiming directly at peptide targets. Google DeepMind and Isomorphic Labs publish AlphaFold 3 (Josh Abramson, Jonas Adler and colleagues, Nature), extending prediction to biomolecular interactions: proteins with DNA, RNA, ligands. In October, the Nobel Prize in Chemistry goes to David Baker for computational protein design, and jointly to Demis Hassabis and John Jumper for protein structure prediction.

Vázquez Torres S. et al., Nature 626, 435 (2024) — abstract · Abramson J. et al., Nature 630, 493 (2024) — abstract · Nobel Prize in Chemistry 2024

2025The peptide-discovery boom

Reviews start mapping an industry: AI-driven antimicrobial peptide discovery — mining genomes for candidates and generating new ones — becomes its own literature (Szymczak, Zarzecki, Wang & Duan, Accounts of Chemical Research). The bottleneck moves from "can AI design a peptide?" to "can a person get a straight, evidence-grounded answer about one?"

Szymczak P. et al., Acc Chem Res 58, 1831 (2025) — abstract