Convenience wrapper that simulates pseudobulks from single-cell data, processes simulated and real bulk data, trains the TAPE autoencoder, and predicts cell-type fractions with optional adaptive refinement.
Usage
tape(
sc_data,
real_bulk,
variance_threshold = 0.98,
scaler = "mms",
d_prior = NULL,
mode = "overall",
adaptive = TRUE,
sparse = TRUE,
batch_size = 128L,
epochs = 128L,
seed = 0L,
samplenum = 5000L,
n = 500L,
celltype_col = "CellType",
assay = "RNA",
slot = "counts"
)Arguments
- sc_data
A `Seurat` object, a `SingleCellExperiment` object, or a matrix/data.frame with cells in rows and genes in columns.
- real_bulk
Numeric matrix with genes in rows and samples in columns.
- variance_threshold
Numeric scalar in `[0, 1]` used for variance-based gene filtering.
- scaler
Character scalar. Either `"mms"` or `"ss"`.
- d_prior
Numeric vector or `NULL`. Dirichlet prior used in simulation.
- mode
Character scalar. Either `"overall"` or `"high-resolution"`.
- adaptive
Logical scalar. Whether to use adaptive refinement.
- sparse
Logical scalar. Whether to simulate sparse mixtures.
- batch_size
Integer scalar. Batch size for training.
- epochs
Integer scalar. Number of training epochs.
- seed
Integer scalar. Random seed.
- samplenum
Integer scalar. Number of simulated pseudobulk samples.
- n
Integer scalar. Number of cells per simulated pseudobulk.
- celltype_col
Character scalar. Metadata column containing cell-type labels.
- assay
Character scalar giving the assay name for `Seurat` input.
- slot
Character scalar giving the assay slot or assay name to extract.
Value
A named list with components:
- sigm
Predicted signature matrix output, depending on `mode` and `adaptive`.
- pred
Predicted cell-type proportions with samples in rows and cell types in columns.
Details
This function provides an R implementation of TAPE (Tissue-AdaPtive autoEncoder) as described by Chen et al. (2022), using torch for model training, adaptive refinement, and prediction. The implementation follows the TAPE methodology for pseudobulk simulation, preprocessing, autoencoder training, and optional tissue-adaptive prediction within an R-based workflow.
The model architecture and workflow are implemented in torch in R following the PyTorch implementation provided in the original TAPE repository.
References
Chen, Y., Wang, Y., Chen, Y., Cheng, Y., Wei, Y., Li, Y., Wang, J., Wei, Y., Chan, T.-F., & Li, Y. (2022). Deep autoencoder for interpretable tissue-adaptive deconvolution and cell-type-specific gene analysis. Nature Communications, 13(1), 6735.
Original TAPE software repository: https://github.com/poseidonchan/TAPE