Mutation resistance guide
How to prepare inputs, follow persistent tasks and interpret Boltz ensemble comparisons.
Paste one canonical amino-acid sequence for the binder and the exact antigen construct. Mutation positions are 1-based against the pasted antigen—not UniProt numbering unless the constructs are identical.
Add each replacement residue. Duplicate, out-of-range and no-op mutations are rejected before compute is submitted.
The assessment is saved before provider work starts. Select it from the recent-assessments strip or the global Runs & Candidates list to restore the exact task, refresh WT and mutant progress, and reopen collected artifacts. Repeated network retries are protected by an idempotency key.
Each row summarizes all returned models with median and range. A “consistent predicted reduction” requires every mutant model to be below every WT model; overlapping ensembles remain indeterminate.
ipTM, ipSAE, pTM and pLDDT are structural-confidence signals. They are not affinity, ΔΔG, neutralization or clinical resistance measurements. Use the module for triage and confirm important findings experimentally.
Select a completed WT or mutant row to open its integrated 3D structure. Download the representative PDB and complete metrics CSV for audit or downstream analysis. “Use inputs again” prepares a new run but never silently resubmits an old one.