Experimental Features

The Experimental section of the web app provides tools for quick, one-off tasks — useful for testing new features.

Caution

Experimental features are under active development. Their interfaces and outputs may change without notice.

Note

The Experimental section is entitlement-gated: it only appears on the dashboard for accounts that have been granted access. If you cannot see it, ask your administrator.

Single Epitope Prediction

Predicts BCR epitope clusters for a single antibody using a trained classifier.

Usage

1. Select a Model version from the dropdown (current: model_v1.2)
2. Enter the Heavy chain sequence (required) — full amino acid sequence
3. Optionally enter a Light chain sequence to improve prediction accuracy
4. Set Top clusters to report (default: 10)
5. Set Min probability threshold (recommended: 0.5)
6. Click Predict Epitope Clusters

Result columns

Column	Description
rank	Rank for this query (1 = highest probability)
mean_prob	Mean classifier probability across all reference antibodies in this cluster
msa_cluster_id	Fine-grained epitope cluster ID (MSA-based)
epitope_cluster	Broader epitope cluster grouping
ag_cluster_id	Antigen cluster ID
antigen	Antigen name associated with the predicted cluster
n_ref_abs	Number of reference antibodies in this cluster
rep_ids	Top-3 representative reference antibody IDs

---

Single Sequence Search

Search a single amino acid sequence against one or more reference databases using nearest-neighbor retrieval. Chain type and species are auto-detected from the input sequence.

Usage

1. Enter an amino acid Query sequence — chain type and species are detected automatically
2. Select one or more Databases to search (default: SAbDab). Once the chain type is known, the list is filtered to the matching receptor class — BCR databases for IGH/IGK/IGL and TCR databases for TRA/TRB/TRD/TRG. All databases stay available until a chain type is determined.
3. Set Nearest neighbors to retrieve — candidates fetched from the indexes (default: 1000)
4. Set Results to keep — top hits returned after scoring (default: 100)
5. Choose Search mode: Full paratope (default) or CDR3 only
6. Click Search

Results can be downloaded as CSV.

Result columns

Column	Description
database	Source database for this match
aligned_sequence	Pseudo-sequence alignment of the matched target
distance	Raw levenshtein distance — lower is more similar
sequence_identity	Fraction of matching residues at non-gap positions (0–1)
coverage	Fraction of positions where both sequences have a residue (0–1)
blosum62_score	Mean per-position BLOSUM62 score — higher is more similar
nw_score	Needleman–Wunsch score normalised by the longer sequence — higher is more similar
cdr3_length_diff	CDR3 length difference (query − target; 0 = equal length)
deamidation	Count of deamidation motifs (NG, NS) in the padded CDR regions
isomerization	Count of isomerization motifs (DG, DS) in the padded CDR regions
oxidation	Count of oxidation-prone residues (M, W) in the padded CDR regions
refined	Whether CDR3 gap positions were repositioned via BLOSUM62 sliding-window refinement
glycosylation	Potential N-linked glycosylation sequons (N-X-S/T, X ≠ Pro) scanned across the full target sequence — count and 1-based positions
dataset_name	Source dataset the target sequence belongs to
condition	Condition label associated with the target sequence
subject_idx	Subject identifier within the source dataset
extra columns	Database-specific metadata (e.g. antigen.epitope, v.alpha) — fields listed under Database Info

Note

Sequence-liability annotations. glycosylation is scanned across the full target amino-acid sequence. The chemical liabilities — deamidation, isomerization, and oxidation — are scanned only on the padded CDR regions (IMGT 24–41, 53–68, 102–120): each hypervariable loop plus 3 flanking, solvent-exposed framework (Vernier-zone) residues, ignoring the buried core. All four are blank when the database stores no amino-acid sequence for the target.

---

AF3 Config Builder

Builds an AlphaFold3 job config for an antibody–antigen complex, so you do not have to assemble the JSON by hand.

Supply the antigen as a UniProt accession — the sequence is fetched for you — together with the antibody heavy and light chain sequences. The tool assembles the corresponding AlphaFold3 input config, which you can then copy out and run wherever you execute AlphaFold3.

Note

This tool is embedded from a separate MyImmune service and opens full-screen inside the app. It only builds the config — it does not run the folding job, and nothing is saved to your account.

---

When to use vs. full pipelines

Scenario	Use
Testing one sequence	Experimental tools
Analyzing a full cohort (100+ sequences)	Bulk pipeline jobs
Need results saved on cloud	Bulk pipeline jobs
Quick sanity check before a batch run	Experimental tools