The Protein Structure Prediction Revolution: AlphaFold’s Evolution from Research Tool to Drug Discovery Platform

Proteins are life’s primary functional executors — enzymes, antibodies, receptors, and structural proteins are all proteins. Protein function is determined by 3D structure, which is formed by amino acid sequences folding according to complex physicochemical rules. The “protein folding problem” — predicting structure from sequence — was molecular biology’s most important unsolved mystery for 50 years. AlphaFold2 essentially solved it.

## AlphaFold Progress

**AlphaFold2 (2020, DeepMind)**: at CASP14 (protein structure prediction competition), dramatically surpassed all competing approaches with 90+ GDT_TS scores (near-atomic precision), shocking the scientific community. Technical core: attention mechanism-based neural network using multiple sequence alignments (MSA) and pairwise sequence representations to learn evolutionary relationships and infer 3D structural constraints. Open database: ~200 million protein structure predictions, freely accessible to researchers.

**AlphaFold3 (2024, DeepMind/Isomorphic Labs)**: extended from proteins to protein-DNA, protein-RNA, and protein-small molecule complex structure prediction — directly predicting drug molecule-target protein binding modes (docking). This has direct drug discovery value: predicting candidate drug binding pockets and modes to guide molecular design.

## Impact on Drug Discovery

**Structure-based drug design (SBDD)**: designing small molecules that precisely bind active pockets using target protein 3D structures as a basis. AlphaFold provided structures for virtually all human proteins, freeing SBDD from X-ray crystallography and cryo-EM experimental speed limits (typically months to years).

**Orphan target development**: many disease-related proteins lacking experimental structures (orphan targets) can now use AlphaFold-predicted structures as new drug design starting points.

**Protein engineering**: combining AlphaFold with generative protein design tools (RFDiffusion, ProteinMPNN) achieves “design proteins with target function from scratch” — novel enzymes, antibodies, and protein therapeutics beyond natural evolution.

See [AI Drug Discovery](https://sunqi.org/ai-drug-discovery-en/) and [AlphaFold database](https://alphafold.ebi.ac.uk/).