What researchers are trying
AlphaProteo conditions generation on a target structure and preferred binding region, then uses a computational filtering pipeline before laboratory testing.
Prospective de novo binder generation
Can AI generate high-affinity protein binders for diverse targets with fewer rounds of experimental optimization?
Why it matters
Reliable one-pass binder generation could accelerate research reagents, diagnostics and therapeutic lead creation.
Evidence boundary
Performance varies by target; the system did not succeed on every attempted target and is not publicly released for unrestricted reproduction.
Organizations represented
Google DeepMind
Demonstrated evidence
What has actually been shown.
- Prospective wet-lab testing across seven targets.
- Developer-reported higher experimental success and 3–300-fold stronger affinities than selected existing methods.
Unresolved questions
- Which target properties predict designability?
- How transferable are success rates to membrane and flexible targets?
- Can developability, immunogenicity and function be optimized jointly?
Signals to watch next
- Independent access
- Expanded target panels
- Therapeutic-format validation
- Multi-objective design
Connected evidence graph
Related BioAtlas model passports.
AlphaProteo
High-affinity binder generation from DeepMind.
2/7 evidence fields documentedRFdiffusion
Diffusion models that hallucinate brand-new proteins.
5/7 evidence fields documentedProteinMPNN / LigandMPNN
Given a shape, design the sequence that folds into it.
3/7 evidence fields documentedESM3
A generative model that reasons over sequence, structure & function at once.
4/7 evidence fields documentedPrimary and evaluation sources