Dynamo Geneid Convert

Dynamo Gene ID Convert

Goal

Convert Ensembl-style identifiers to gene symbols in a way that another agent can actually rerun: choose the correct conversion path, preserve traceability columns, handle species and version-suffix edge cases, and only hand off to Preprocessor after IDs are standardized.

Quick Workflow

1. Inspect adata.var_names or the raw ID list and determine whether the user needs a batch table or an in-place AnnData update.
2. Strip version suffixes such as .1 into a query column before you treat mapping quality as final.
3. Use convert2gene_symbol(...) when you need an explicit mapping table, reproducible merge logic, or non-human species control.
4. Use convert2symbol(adata, ...) only when in-place AnnData mutation is the right abstraction and the prefix / scopes rules are satisfied.
5. Keep the original identifier in adata.var, set adata.var_names from symbol only after validating duplicates and missing mappings, and run preprocessing afterward, not before.

Interface Summary

• convert2gene_symbol(input_names, scopes='ensembl.gene', ensembl_release=None, species=None, force_rebuild=False) returns a DataFrame indexed by query with at least symbol and _score.
• The live source uses vendored pyensembl data access, not a remote MyGene batch API.
• convert2symbol(adata, scopes=None, subset=True) updates adata.var, adds query plus symbol, and can rewrite adata.var_names in place.
• Preprocessor.preprocess_adata(adata, recipe='monocle', tkey=None, experiment_type=None) is only the downstream handoff. Current source exposes five recipe branches: monocle, seurat, sctransform, pearson_residuals, monocle_pearson_residuals.

Read references/source-grounding.md before documenting parameters more narrowly than the notebook does.

Conversion Path Selection

• Use convert2gene_symbol(...) as the default for notebook conversion work. It is the safest path when you need explicit species, ensembl_release, manual merge logic, or per-ID validation.
• Use convert2symbol(adata) for human or mouse Ensembl-style adata.var_names when direct in-place conversion is acceptable.
• Use convert2symbol(adata, scopes='ensembl.gene') for zebrafish or other non-human prefixes if you still want the in-place helper.
• Do not trust the notebook's generic scopes explanation alone. In current source, convert2gene_symbol keeps scopes only for compatibility, while convert2symbol still branches on scopes and prefix heuristics.

Minimal Execution Patterns

For an explicit mapping table and controlled merge:

import dynamo as dyn

adata = dyn.sample_data.hematopoiesis_raw()
adata.var["ensembl_id"] = adata.var_names
adata.var["query"] = adata.var_names.str.split(".").str[0]

mapping = dyn.preprocessing.convert2gene_symbol(
    adata.var["query"].tolist(),
    species="human",
)

adata.var = (
    adata.var
    .merge(mapping, left_on="query", right_index=True, how="left")
    .set_index(adata.var.index)
)

mapped = adata.var["symbol"].notna()
adata = adata[:, mapped].copy()
adata.var_names = adata.var["symbol"].astype(str)

For in-place conversion on supported prefixes:

import dynamo as dyn

adata = dyn.sample_data.hematopoiesis_raw()
adata.var["ensembl_id"] = adata.var_names
dyn.preprocessing.convert2symbol(adata, subset=True)

For zebrafish IDs with version suffixes:

import dynamo as dyn

result = dyn.preprocessing.convert2gene_symbol(
    ["ENSDARG00000035558.1"],
    ensembl_release=77,
)

# Or, if the user wants in-place AnnData mutation:
dyn.preprocessing.convert2symbol(adata, scopes="ensembl.gene")

Optional Preprocess Integration

• Perform gene-ID conversion before preprocessing unless the user explicitly wants to preserve notebook timing.
• If the user proceeds into preprocessing, default to recipe='monocle' unless they ask for a different branch.
• If the user requests Pearson residuals but still cares about downstream velocity-safe layers, prefer monocle_pearson_residuals.

Read references/preprocess-handoff.md before choosing a non-default recipe.

Validation

After conversion, check these items:

• adata.var["query"] stores the version-stripped identifier actually used for lookup.
• adata.var["symbol"] exists and the mapping rate is acceptable for the dataset.
• The original ID remains preserved, for example in adata.var["ensembl_id"].
• adata.var_names contains symbols only after duplicate and null handling is explicit.
• Representative conversions still match live source behavior: ENSG00000141510 -> TP53 and ENSDARG00000035558(.1) -> gps2 with ensembl_release=77.
• If preprocessing follows, Preprocessor.preprocess_adata(..., recipe=...) should run only after symbol assignment is settled.

Constraints

• Do not describe conversion as MyGene-backed just because older prose or notebook wording suggests that pattern; current source uses vendored pyensembl.
• Do not assume convert2symbol(adata) auto-detects every species. In current source it auto-classifies some human / mouse gene or transcript prefixes, but zebrafish without explicit scopes raises.
• Do not assume the docstring's ensembl_release default is authoritative. Current code assigns 77 when the argument is omitted.
• The first conversion run may install polars or download / index Ensembl data, so cold-start execution can be slower than the notebook suggests.

Resource Map

• Read references/source-grounding.md for inspected signatures, live-source behavior, and branch coverage.
• Read references/conversion-paths.md for human vs zebrafish decision rules and scopes handling.
• Read references/preprocess-handoff.md for the downstream recipe branches.
• Read references/source-notebook-map.md to see which notebook sections were preserved or intentionally dropped.
• Read references/compatibility.md when notebook wording and current source behavior appear to disagree.