CAR Interface Assessment Walkthrough
BetaA practical guide to generating and reviewing isolated-arm structural hypotheses for a dual-target CAR design.
For CD19-BCMA or similar dual-target CAR concepts, this page is a Phase 1 pairwise structural-triage surface. It creates four independent Interface jobs: each isolated binder arm against each antigen. It does not validate fused-chain folding, membrane geometry, binding affinity, or specificity.
Treat pass / review / fail as labels from structural pose heuristics. ipTM, pDockQ, BSA, contacts, and clashes describe generated models; none is direct evidence of affinity, binding, or specificity.
A dual-target CAR binder is often one fused binder-linker-binder chain. Linker length and orientation can change folding, domain packing, and binding-face exposure. The current 2 x 2 mode cannot assess those effects. A single unconstrained sequence-only fold is also not a validated construct-geometry gate; this requires a separate, explicitly annotated research protocol.
Future research mode
Annotate both domain boundaries and linker, then use justified templates/restraints and multiple seeds or samples. Do not infer arm identity from one raw sequence.
Geometry check
Compare the pose ensemble for linker reach, domain separation, clashes, binder-binder packing, exposed binding faces, and convergence.
Phase 1
Use the current 2 x 2 matrix only for isolated-arm pairwise hypotheses, while keeping fused-construct geometry explicitly unresolved.
| Step | Action | What the platform does |
|---|---|---|
| 1 | Paste two isolated CAR binding arms and two antigen ECD sequences. | Inputs are normalized and checked for duplicated arms or probable full-tandem misuse before compute is submitted. |
| 2 | Run the 2 x 2 arm-vs-antigen matrix. | The module creates four child Interface jobs: two intended pairs and two cross-target counter-screens. |
| 3 | Review intended-pair and counter-screen pose support together. | The parent heuristic is pass / review / fail, with ipTM, pDockQ, BSA, clashes, notes, PDB, CSV and JSON artifacts. These are pose signals, not binding measurements. |
| 4 | Open child jobs for structural inspection when a row looks suspicious. | Use the child Interface view for detailed contacts, paratope, epitope footprint and downloadable files. |
Strong intended-pair pose signal + weak counter-screen signal
Most coherent modeled pattern; inspect interfaces and seek orthogonal or experimental evidence.
Strong intended-pair and counter-screen pose signals
The model generated plausible interfaces for both; inspect whether the counter-screen pose is biologically credible before inferring specificity risk.
Weak intended-pair and counter-screen pose signals
The prediction is not discriminative; do not infer weak affinity or specificity from absence of a confident pose.
Weak intended-pair signal + strong counter-screen signal
Treat as a model flag requiring structure review and independent evidence, not as proof of binding failure or cross-binding.
Current scope
Sequence-to-complex triage for four arm-antigen pairs, with persistent history and downloadable artifacts.
Missing before real use
A fused single-chain construct precheck is needed before using pairwise docking as evidence.
Next step
Build Phase 0: fused construct input, full-chain fold, linker/domain plausibility checks, then hand off to this 2 x 2 matrix.