Perform Scissor Screening Analysis

Identifies phenotype-associated cell subpopulations in single-cell data using regularized regression on matched bulk expression profiles. Scissor integrates bulk and single-cell RNA-seq data to identify cells that are significantly associated with phenotypic outcomes.

Usage

DoScissor(
   matched_bulk,
   sc_data,
   phenotype,
   label_type = "scissor",
   alpha = c(0.05, NULL),
   cutoff = 0.2,
   family = c("gaussian", "binomial", "cox"),
   reliability_test = list(
     run = FALSE, # whether to run reliability test
     n = 10L, # permutation times
     nfold = 10L # cross validation folds
   ),
   cell_evaluation = list(
     run = FALSE, # whether to run cell evaluation
     benchmark_data = "path_to_file.RData", # path to benchmark data
     FDR_cutoff = 0.05,
     bootstrap_n = 100L
   ),
   path2load_scissor_cache = NULL,
   path2save_scissor_inputs = "Scissor_inputs.RData",
   ...
)

Arguments

matched_bulk

Normalized bulk expression matrix (features × samples). Column names must match phenotype identifiers.

sc_data

Seurat object containing single-cell RNA-seq data.

phenotype

Clinical outcome data. Can be: - Vector: named with sample IDs - Data frame: with row names matching bulk columns

label_type

Character specifying phenotype label type (e.g., "SBS1", "time"), stored in scRNA_data@misc

alpha

Parameter used to balance the effect of the l1 norm and the network-based penalties. It can be a number or a searching vector. If alpha = NULL, a default searching vector is used. The range of alpha is between 0 and 1. A larger alpha lays more emphasis on the l1 norm.

cutoff

(default: 0.2). When alpha=NULL, the cutoff is used to determine the optimal alpha. Higher values increase specificity.

family

Model family for outcome type: - "gaussian": Continuous outcomes - "binomial": Binary outcomes (default) - "cox": Survival outcomes

reliability_test

List controlling reliability testing:

run: Whether to perform reliability test (default: FALSE)
n: Permutation times (default: 10L)
nfold: Cross-validation folds (default: 10L)

cell_evaluation

List controlling cell evaluation:

run: Whether to perform cell evaluation (default: FALSE)
benchmark_data: Path to benchmark data (RData file)
FDR_cutoff: FDR threshold for evaluation (default: 0.05)
bootstrap_n: Bootstrap iterations (default: 100L)

path2load_scissor_cache

Path to precomputed Scissor inputs (RData file). If provided, skips recomputation (default: NULL).

path2save_scissor_inputs

Path to save intermediate files (default: "Scissor_inputs.RData").

...

Additional arguments. Currently supports:

verbose: Logical indicating whether to print progress messages. Defaults to TRUE.
seed: For reproducibility, default is 123L
assay: Assay to use for single-cell data. Defaults to `"RNA"

Value

A list containing:

scRNA_data

A Seurat object with screened cells containing metadata:

scissor: "Positive"/"Negative"/"Neutral" classification
label_type: Outcome label used

scissor_result

Raw Scissor results

reliability_result

If reliability_test=TRUE, contains:

statistic: A value between 0 and 1
p: p-value of the test statistic
AUC_test_real: 10 values of AUC for real data
AUC_test_back: A list of AUC for background data

cell_evaluation

If cell_evaluation=TRUE, contains:

evaluation_res: A data.frame with some supporting information for each Scissor selected cell

LICENSE

Licensed under the GNU General Public License version 3 (GPL-3.0). A copy of the license is available at https://www.gnu.org/licenses/gpl-3.0.en.html.

References

Sun D, Guan X, Moran AE, Wu LY, Qian DZ, Schedin P, et al. Identifying phenotype-associated subpopulations by integrating bulk and single-cell sequencing data. Nat Biotechnol. 2022 Apr;40(4):527–38.

Examples

if (FALSE) { # \dontrun{
# Binary outcome example
res <- DoScissor(
  matched_bulk = bulk_matrix,
  sc_data = seurat_obj,
  phenotype = a_named_vector,
  family = "binomial"
)
} # }