Runs OmicsTweezer cell-type deconvolution from a single-cell reference and real bulk expression profiles. The workflow simulates pseudobulks, processes simulated and real bulk data, trains three neural network architectures, and averages their predictions.
Usage
omics_tweezer(
sc_data,
real_bulk,
ot_weight = 1,
scale_minmax = FALSE,
samplenum = 5000L,
n_cells_per_bulk = 500L,
sparse = FALSE,
variance_threshold = 0.98,
scaler = c("ss", "mms", "none"),
batch_size = 128L,
epochs = 30L,
learning_rate = 1e-04,
seed = 2021L,
celltype_col = "CellType",
assay = "RNA",
slot = "counts",
verbose = TRUE,
device = c("auto", "cpu", "cuda"),
cuda_index = NULL
)Arguments
- sc_data
A single-cell reference object. Supported inputs are a `Seurat` object, a `SingleCellExperiment` object, a matrix, or a `data.frame`.
- real_bulk
Numeric matrix or `data.frame` with genes in rows and real bulk samples in columns.
- ot_weight
Numeric scalar. Weight applied to the Wasserstein-style domain adaptation loss.
- scale_minmax
Logical scalar. If `TRUE`, applies an outer global min-max scaling step to simulated and real bulk count matrices before processing.
- samplenum
Integer scalar. Number of simulated pseudobulk samples.
- n_cells_per_bulk
Integer scalar. Target number of single cells per simulated pseudobulk.
- sparse
Logical scalar. If `TRUE`, applies sparse perturbation during pseudobulk simulation.
- variance_threshold
Numeric scalar. Fraction of genes used to define the variance cutoff during processing.
- scaler
Character scalar specifying the per-sample scaling method. One of `"ss"`, `"mms"`, or `"none"`.
- batch_size
Integer scalar. Minibatch size for training and prediction.
- epochs
Integer scalar. Number of training epochs for each architecture.
- learning_rate
Numeric scalar. Learning rate passed to Adam.
- seed
Integer scalar random seed.
- celltype_col
Character scalar giving the metadata column containing cell-type labels for `Seurat` and `SingleCellExperiment` input.
- assay
Character scalar giving the assay name to extract from a `Seurat` object.
- slot
Character scalar giving the assay slot or assay name to extract.
- verbose
Logical scalar. If `TRUE`, prints workflow progress messages.
- device
Character scalar. One of `"auto"`, `"cpu"`, or `"cuda"`. If `"auto"`, CUDA is used when available, otherwise CPU is used.
- cuda_index
Optional integer scalar giving the CUDA device index to use when `device = "cuda"`. CUDA indices are zero-based.
Value
A named list with components:
- pred
Data frame of averaged predicted cell-type proportions.
- per_model
Named list of predictions from the `m256`, `m512`, and `m1024` architectures.
- simudata
Simulated pseudobulk data returned by [omics_simulate()].
- processed
Processed matrices returned by [omics_process()].
- models
Named list of trained OmicsTweezer models.
- loss_history
Data frame containing epoch-level training losses for all architectures.