Number, model and analyse antibodies

What this involves

Working with antibody sequences requires precise structural annotation before any downstream modeling or design can proceed. This task begins with sequence sanitation and numbering according to standard immunological conventions such as IMGT, Chothia, or Kabat. Assigning standardized positions allows you to demarcate framework regions from complementarity-determining regions (CDRs), isolate variable domains, and identify sequence irregularities before committing to structural or developability predictions.

Structural analysis requires determining whether an empirical structure already exists in the Protein Data Bank (PDB) before undertaking de novo prediction. When modeling is necessary, dedicated antibody tools such as ImmuneBuilder (ABodyBuilder2 and NanoBodyBuilder2) or ESMFold generate three-dimensional coordinates. Evaluating these coordinates demands inspection of residue-level confidence (pLDDT) rather than overall averages, allowing you to differentiate well-ordered framework scaffolds from flexible or disordered CDR loops.

Finally, sequence-based developability profiling screens candidates for biochemical vulnerabilities. Deterministic sequence analyses measure properties such as net charge, molecular weight, grand average of hydropathicity (GRAVY), and extinction coefficients under reduced or cystine-bonded states. Specialised models, including MAMMAL solubility for binary solubility prediction, AbLang2 for antibody-specific sequence embeddings, AntiFold for inverse folding, and Sapiens for germline-guided humanisation, can then be executed to guide engineering decisions.

How it goes

1. 1Sequence curation and verificationClean raw sequences of FASTA headers, gaps, and non-standard characters, checking for ambiguity codes, stop codons, and cleaved signal peptides prior to computation.
2. 2Variable domain numbering and annotationExecute ANARCI on the variable domain sequence to assign IMGT, Chothia, or Kabat numbering schemes and define framework and CDR boundaries.
3. 3Structural retrieval or coordinate predictionQuery the PDB via lookup_structure for experimentally resolved entries, falling back to ImmuneBuilder or ESMFold when no experimental coordinates exist.
4. 4Per-residue confidence evaluationExtract per-residue pLDDT metrics from predicted models to segment chains into confidence bands, verifying framework stability (>90 or 70–90) versus flexible loop regions.
5. 5Developability profiling and sequence designCalculate physicochemical properties with analyze_sequence, predict solubility using MAMMAL solubility, or apply AbLang2, AntiFold, or Sapiens for optimization.

Start it here

Ask in plain language. The agent works on a Linux machine of its own — it installs and runs what the job needs, keeps every file, and reads the results back with the numbers explained. Nothing to set up, and a long run continues after you close the page.

What to watch for

• Never report a mean pLDDT alone, as a single chain average obscures whether flexible CDR loops or tails drag down an otherwise well-folded framework.
• Folding an unprocessed chain that retains its N-terminal signal peptide yields an unnatural lone helix that depresses the overall confidence score.
• Do not fold a point mutant to observe its structural change, because the difference between wild-type and mutant predictions is smaller than the model's error margin.
• Selecting the cystine extinction coefficient for reduced samples or the reduced coefficient for disulfide-bonded proteins will corrupt all downstream concentration measurements.
• The instability index is calibrated against in vivo half-life and does not reflect thermal stability or antibody shelf life.

What it runs

Questions