How google-deepmind/alphaprotein-novo works
AlphaProtein Novo designs enzymes by generating protein structure and sequence conditioned on a catalytic motif and ligand, then optionally redesigning sequence, predicting structures, and evaluating designs. The documented entry point is a manifest-driven pipeline that runs these stages and can resume campaigns. The graph includes motif/data preparation and the model’s sampling layer as internal components. Stage links follow the README; source excerpts are partial, so internal links are limited to relationships supported by the SOURCE INDEX or excerpts.
Main entry points: run_pipeline.py Pipeline runner · evaluate_design.py Metrics runner · run_generator.py Design generator · design_manifest.py Manifest and design jobs
Components 20
Campaign orchestration
- Pipeline runner
run_pipeline.py - Design artifacts and reports
Motif-conditioned generation
- Design generator
run_generator.py - Motif specification
src/alphaprotein_novo/data/motif_spec.py - Diffusion sampling
src/alphaprotein_novo/model/sampling.py - Protein denoiser
src/alphaprotein_novo/model/denoiser.py - Model configuration
src/alphaprotein_novo/model/config.py - Generator parameters
src/alphaprotein_novo/model/params.py
Structure and sequence stages
- Sequence redesign
run_ligandmpnn.py - Structure prediction
run_alphafold.py
Design evaluation
- Metrics runner
evaluate_design.py - Structure and enzyme metrics
src/alphaprotein_novo/metrics/structure_metrics.py - Ligand metrics
src/alphaprotein_novo/metrics/ligand_metrics.py - Enzyme-specific metrics
src/alphaprotein_novo/metrics
Shared modeling data
- Manifest and design jobs
src/alphaprotein_novo/data/design_manifest.py - Structure features
src/alphaprotein_novo/data/structure_features.py - Structure utilities
src/alphaprotein_novo/data/structure_utils.py - Ligand data
src/alphaprotein_novo/data/ligand_data.py
Other components
- Design engineer
- AlphaFold 3
Connections 25
- Design engineer → Pipeline runner: configures campaign
README.md - Pipeline runner → Manifest and design jobs: loads settings
README.md - Pipeline runner → Design generator: runs generation
README.md - Pipeline runner → Sequence redesign: optionally redesigns
README.md - Pipeline runner → Structure prediction: runs prediction
README.md - Pipeline runner → Metrics runner: runs evaluation
README.md - Design generator → Manifest and design jobs: reads jobs
run_generator.py - Design generator → Motif specification: uses motif specs
run_generator.py - Design generator → Diffusion sampling: uses sampler
run_generator.py - Diffusion sampling → Protein denoiser: calls denoiser
src/alphaprotein_novo/model/sampling.py - Protein denoiser → Model configuration: uses configuration
src/alphaprotein_novo/model/denoiser.py - Design generator → Generator parameters: loads parameters
run_generator.py - Motif specification → Structure features: uses feature utilities
src/alphaprotein_novo/data/motif_spec.py - Motif specification → Structure utilities: uses structure helpers
src/alphaprotein_novo/data/motif_spec.py - Structure utilities → Ligand data: uses ligand data
src/alphaprotein_novo/data/structure_utils.py - Sequence redesign → Structure utilities: updates structures
run_ligandmpnn.py - Structure prediction → Manifest and design jobs: reads prediction specs
run_alphafold.py - Structure prediction → AlphaFold 3: invokes folding
README.md - Metrics runner → Manifest and design jobs: loads design settings
evaluate_design.py - Metrics runner → Structure and enzyme metrics: computes metrics
evaluate_design.py - Metrics runner → Ligand metrics: computes ligand scores
evaluate_design.py - Metrics runner → Enzyme-specific metrics: selects evaluation suite
README.md - Structure and enzyme metrics → Ligand metrics: uses ligand metrics
src/alphaprotein_novo/metrics/structure_metrics.py - Metrics runner → Design artifacts and reports: writes reports
- Pipeline runner → Design artifacts and reports: organizes stage artifacts
README.md