SCmaize Apps

Job Status

Job	App	Status	Created	Result

SCmaize Apps User Manual

Use this guide to prepare input files, download examples, understand job retention, and interpret output files.

Data retention and privacy

Important Inputs and results are retained for 14 days only. Keep the private status link to retrieve results.

Uploaded files are used only to run the requested job.
Uploaded files are not used to retrain or improve the scMaize model.
Anyone with the private status link can access that job result during the retention window.

General input requirements

File-based tools accept AnnData `.h5ad` files with cells in `obs` and genes in `var` / `var_names`.
Gene identifiers should be maize gene symbols or stable maize gene IDs that can match the scMaize 15,000-gene vocabulary.
Expression values should be a cell-by-gene matrix in `.X`; log-normalized maize single-cell RNA-seq is recommended for demos.
Only maize data is currently supported. Other species are not expected to match the model vocabulary reliably.
Uploads are limited to 100 MB, and each submitted dataset is capped at 20,000 cells on the compute side.

Expected AnnData layout .X: cells x genes expression matrix .obs: one row per cell .var_names: maize gene symbols or stable IDs, for example Zm00001d021668

Cell Type Annotation

Input a query `.h5ad` file. Optionally provide a custom reference `.h5ad` with `obs` columns named `tissue` and `cell_type`; otherwise the production scMaize atlas reference is used.

Use `Tissue restriction` when the query is known to come from one tissue, for example `Root`. In advanced options, `Neighbors` controls how many nearest reference cells vote during label transfer. The default is 30.

Output is a CSV with `predicted_cell_type`, `predicted_tissue`, and `confidence`. Confidence summarizes agreement among nearest reference neighbors, not an absolute probability.

Try this example Query: test_query_multi.h5ad Reference: test_reference_multi.h5ad Neighbors: 30 Result: scmaize_annotations.csv

test_query_multi.h5adDownload query test_reference_multi.h5adDownload reference eval_reference.h5adDownload larger reference eval_root_qry.h5adDownload root query eval_root_ref.h5adDownload root reference

Cell Embeddings

Input a query `.h5ad` file. `standard` mode returns CLS cell embeddings. `dualStream` combines CLS embeddings with PCA expression components for a richer cell-state representation.

Output is a NumPy `.npy` matrix with one row per cell. In `standard` mode with default settings, `test_query_multi.h5ad` produces a `100 x 384` matrix. In `dualStream` with `Embedding dim = 128` and `PCA dim = 64`, it produces a `100 x 192` matrix.

Try this example Query: test_query_multi.h5ad Mode: dualStream Embedding dim: 128 PCA dim: 64 Result: scmaize_embeddings.npy

test_query_multi.h5adDownload query

Expression Imputation

Input a query `.h5ad` file. The tool uses gene-neighbor relationships from scMaize gene embeddings to estimate sparse or missing expression values.

`Neighbors` controls how many similar genes contribute to each imputed value. The default is 10. Output is an imputed `.h5ad` matrix with the same cell and gene axes as the input.

Try this example Query: test_query_multi.h5ad Model: scMaizeGO Neighbors: 10 Result: scmaize_imputed.h5ad

test_query_multi.h5adDownload query

Gene Similarity

No `.h5ad` file is required. Enter comma-separated maize gene symbols or stable IDs.

Output is a ranked CSV table with the query gene, matched similar gene, rank, and similarity score. Similarity is computed in scMaize gene embedding space and should be interpreted as functional relatedness.

Try this example Genes: CYP704B1, Zm00001d021668 Top N: 10 Model: scMaizeGO Result: scmaize_similarity.csv

Example file notes

The example `.h5ad` files are lightweight maize single-cell RNA-seq datasets for smoke testing and format reference. They include multi-tissue query/reference files and Root-only query/reference files. The small bundled reference file is an example template for custom reference format, not the production annotation reference used by default.