Miniprotein
rbx1_v2_0556
GLHAIRDEVNAVRAKLHNMETKLDQMKHRITTARNDVYLMKVKLFIVRFEMFLMRVDVSTIRAEVADLESRATSFKNRET
No experimental data
This protein hasn't been validated in the lab yet.
This protein was designed using TTHybrid Boltz-ensemble structural cores + pipeline-composite cores + apex variant
Generation & pre-filtering: Candidates were produced with our in-house TRAIRC mini-binder workflow: physics-informed scoring and length/sequence constraints during a generative/search stage, plus first-pass foldability metrics and ESM-2–based per-residue log-likelihood (normalized within the candidate pool). No LLM “prompt” was used for final sequence selection; rankings used numerical scores only.
Boltz-2 structural evaluation: Each candidate binder was modeled as a complex against four distinct RBX-1–related structural contexts (four separate inputs per design). We recorded Boltz-2 outputs including complex ipTM, global confidence, and interface iPDE. Ensemble metrics are the arithmetic mean of those four runs. Inference used the same Boltz-2 setup for all jobs in this batch (consistent checkpoint and sampling configuration across the study).
Score normalization: Within the Boltz-scored pool, mean ipTM, mean confidence, and mean iPDE were min–max normalized to [0,1]; iPDE was inverted so lower (better) values map to higher scores.
Structural-only rank (22 cores): Composite = 0.6·ipTM_norm + 0.2·confidence_norm + 0.2·iPDE_norm (ensemble means).
Full pipeline rank (8 cores): Composite = 0.30·ipTM_norm + 0.10·confidence_norm + 0.10·iPDE_norm + 0.20·physics_norm + 0.15·foldability_norm + 0.15·ESM2_norm, with the same per-metric normalization scheme as above and 0.5 neutral weight for missing ESM2_norm. Pipeline slots were filled in descending composite order among designs not in the structural top-22, requiring mean ensemble ipTM ≥ 0.38; if fewer than eight qualified, remaining slots were filled by the next-best structural ranks.
Variants (70): Single amino-acid substitutions along the top structural core sequence; proline and cysteine were excluded as replacement residues; random seed = 42 for reproducible substitution placement without duplicate position/residue pairs.
Final portfolio: 22 structural + 8 pipeline + 70 variants = 100 sequences.