Skip to main content
Consensus Operator Specificity
License: This constraint can use multiple tools, each under its own license. See the Tools Used tab and each tool’s page for license details.

This constraint is open source. Any third-party models, product names, or trademarks referenced are the property of their respective owners, and Proto is not affiliated with them.

evo-design/proto-language/proto_language/constraint/protein_structure/consensus_specificity_constraint.py
View source
Constraint contributors
Consensus operator-readout specificity score (C = z_NA-MPNN + z_DeepPBS). Predicts (or reuses) a protein-DNA complex for each candidate, runs BOTH NA-MPNN and DeepPBS readout models, computes each model’s dyad-aware sliding-window specificity margin (target vs. best off-target), z-normalizes each margin against the shipped reference distribution, and combines them as C = z_NA-MPNN + z_DeepPBS. DeepPBS failures degrade to z_DeepPBS = 0.

​
API Reference

ConfigConsensusSpecificityConfig Source
Config for the consensus-operator-specificity constraint.
​
string
required
Target DNA motif in A/C/G/T alphabet.
​
List[string]
required
Off-target motifs used for the specificity margin.
​
List[integer]
required
0-based DNA indices mapped to motif positions.
​
number
default:"1.0"
Sliding-logprob margin scale (kept for parity and metadata).
​
integer
default:"0"
Canonical DNA chain label for motif indexing when multiple DNA chains exist (fwd strand 0).
​
string
Path to JSON reference margin mean/std for NA-MPNN and DeepPBS.
​
NAMPNNSpecificityConfig
Tool config for na-mpnn-specificity predictions.
​
DeepPBSSpecificityConfig
Tool config for deeppbs-specificity predictions.
​
enum
default:"alphafold3"
Predictor for the protein-DNA complex; must be DNA-capable (alphafold3/boltz2/protenix).Options: esmfold, esmfold2, alphafold3, boltz2, chai1, protenix, alphafold2, alphafold2_binder
​
ESMFoldConfig
Configuration for ESMFold structure prediction.
​
ESMFold2Config
Configuration for ESMFold2 structure prediction.
​
AlphaFold3Config
Configuration for AlphaFold3 structure prediction.
​
Boltz2Config
Configuration for Boltz2 structure prediction.
​
Chai1Config
Configuration for Chai1 structure prediction.
​
ProtenixConfig
Configuration for Protenix structure prediction.
​
AlphaFold2Config
Configuration for the general AlphaFold2 multimer structure predictor.
​
AlphaFold2BinderStructureConfig
Configuration for the AF2 binder-design backend.
ReturnsConstraintOutput
One result per candidate; lower score is better. Score is 0.5 - 0.5 * tanh(C / 2) so a large positive consensus (strong discrimination) maps toward 0.0. Metadata carries each model’s margin, z-score, and the consensus C.

​
Usage

python

​
Metadata