Support vdj modality for DMS datasets

[mzueva]

Greptile Summary

This PR recognizes producer-declared VDJ modality when constructing embedding scopes, introduces an explicit unknown-receptor state, broadens model selection for DMS VDJ data, and updates Platforma SDK dependencies.

• Modality — the producer-declared run category stored under pl7.app/modality; the PR detects "vdj" and uses it to distinguish V-domain sequences from ordinary amplicon/peptide sequences.
• Scope feature — the biological sequence region used for model compatibility; VDJ amplicon-sequence columns now become VDJRegion, while VDJ CDR3 region columns become CDR3.
• Scope receptor — the receptor associated with a selectable scope; the new ScopeReceptor type extends WorkflowReceptor with "unknown" when VDJ data supplies neither receptor nor chain metadata.
• Model compatibility — filtering that determines which embedding models support a scope; unknown-receptor scopes now bypass receptor and heavy-chain specialization gates, while ESM-2 remains the default.
• Sequence selector — metadata criteria used to discover input columns; a selector was added for amino-acid CDR3 columns represented as pl7.app/sequence.
• Recommended model — the initial model selected for a scope; unknown-receptor scopes now default to universal ESM-2 instead of a receptor-specific specialist.

Confidence Score: 5/5

The PR appears safe to merge because no concrete changed-code defect or actionable security issue remains.

The new modality classification, receptor representation, compatibility filtering, and dependency refresh are internally consistent, and the potentially surprising model-selection behavior is explicitly documented as intentional.

Important Files Changed

Filename	Overview
model/src/scopes.ts	Adds VDJ modality detection, CDR3 discovery, unknown-receptor resolution, and modality-sensitive scope classification.
model/src/compat.ts	Widens compatibility APIs to ScopeReceptor, relaxes specialization gates for unknown receptors, and defaults those scopes to ESM-2.
model/src/index.ts	Reads entity-axis modality metadata and passes it into scope construction while exporting the new receptor type.
model/src/types.ts	Introduces ScopeReceptor and propagates it through persisted selected scopes and generated scope configuration.
pnpm-workspace.yaml	Updates the cataloged Platforma SDK patch versions used across workspace packages.
pnpm-lock.yaml	Refreshes resolved Platforma SDK packages and their transitive dependency graph consistently with the workspace catalog.
.changeset/vdj-modality-model-selection.md	Documents the new VDJ scope classification, unknown-receptor behavior, model-selection tradeoff, and package release impact.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Input entity-axis metadata] --> B{pl7.app/modality is vdj?}
  B -- No --> C[Classify amplicon sequence as peptide]
  B -- Yes --> D[Classify whole variant as VDJRegion]
  B -- Yes --> E[Classify region column as CDR3]
  A --> F{Receptor or chain supplied?}
  F -- Yes --> G[Resolve IG, TCRAB, or TCRGD]
  F -- No, VDJ --> H[ScopeReceptor unknown]
  G --> I[Apply receptor and heavy-chain model gates]
  H --> J[Offer feature-compatible VDJ models]
  J --> K[Default to ESM-2]
  C --> L[Build selectable scope configuration]
  D --> L
  E --> L
  I --> L
  K --> L

Loading

_{Reviews (1): Last reviewed commit: "Support vdj modality for DMS datasets" | Re-trigger Greptile}

Context used:

• Context used - Terms is a types in codebase. Provide the list of ... (source)