AlphaProteo
DeepMind's binder-design system that produced picomolar binders, including the first AI-designed VEGF-A binder.
About
AlphaProteo, introduced by DeepMind in September 2024, is a machine-learning system for designing novel high-affinity protein binders. It pairs a generative model that proposes binder candidates for a given target with a filtering model that predicts which candidates will succeed experimentally. The system was trained on the Protein Data Bank plus more than 100 million AlphaFold-predicted structures.
The technical report describes wet-lab testing against eight structurally diverse, therapeutically relevant targets. For seven of them, AlphaProteo achieved experimental hit rates of 9 to 88 percent and binding affinities 3 to 300 times better than the best existing methods, including RFdiffusion, reaching picomolar to low-nanomolar affinities after a single round of medium-throughput screening with no affinity maturation. It produced the first AI-designed binder for VEGF-A, showed VEGF signaling inhibition in human cells and SARS-CoV-2 neutralization, and had its designed complexes confirmed by cryo-EM and X-ray crystallography. One target failed completely. No code, weights, or API have been released.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (Zambaldi et al., arXiv 2409.08022, 2024): 8 targets tested, 7 with 9-88% experimental hit rates and 3-300x better affinities than the best prior methods; 80-960 pM affinities on four targets with no affinity maturation; VEGF signaling inhibition and SARS-CoV-2 neutralization in cell assays; structures confirmed by cryo-EM and X-ray
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: DeepMind technical report arXiv 2409.08022 and the DeepMind research blog confirm the system design, the eight-target wet-lab campaign, and the 3-300x affinity improvements; the report itself discloses the one failed target. Strengths: unusually thorough experimental validation, strongest reported binder-design results to date, honest failure reporting. Docked for: (1) fully closed — no weights, code, or API, so no claim can be independently tested; (2) requires an experimental target structure and a specified binding site; (3) failed entirely on one of eight targets, so generalization is unproven. License: proprietary; nothing released.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 1.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Strongest reported binder-design results to date with unusually thorough wet-lab validation (cryo-EM, X-ray).
- · Fully closed: no claim can be independently tested.
- · Failed entirely on one of eight targets; requires an experimental target structure.