Perform SIDISH Screening Analysis

Usage

DoSIDISH(
  matched_bulk,
  sc_data,
  phenotype,
  label_type = "SIDISH",
  phenotype_class = "survival",
  sidish_params = list(),
  ...
)

matched_bulk

Matrix or data frame of preprocessed bulk RNA-seq expression data (genes x samples). Column names must match names/IDs in phenotype.

sc_data

A Seurat object containing scRNA-seq data to be screened.

phenotype

Phenotype data, either: - Patient survival Data frame with row names matching matched_bulk columns, colnames named "time" and "status"

label_type

Character specifying phenotype label type

phenotype_class

Type of phenotypic outcome (must be consistent with input data): - "survival": Survival infomation

sidish_params

List of SIDISH algorithm parameters including: Preprocessing parameters:

patient_id: column name for patient identifier in metadata (default: "Sample")
celltype_name: column name for cell type annotation in metadata (default: "celltype_major")
processed: whether input data is already preprocessed (default: TRUE)
n_genes_by_counts: minimum number of genes expressed per cell filter threshold (default: 5000)
pct_counts_mt: maximum percentage of mitochondrial genes filter threshold (default: 10)
batch_correction: whether to perform batch correction (default: FALSE)
survival_: column name for survival time in phenotype data (default: "time")
status: column name for event status in phenotype data (default: "status")

Execution environment:

device: computation device, "cuda" for GPU acceleration or "cpu" for CPU-only (default: "cuda")
use_spatial_graph: whether to use spatial graph information (default: FALSE)
k_neighbors: number of neighbors for graph construction (default: NULL, auto-detected)

Phase 1: VAE training parameters:

phase1_epochs: total epochs for VAE training (default: 225)
phase1_i_epochs: interval epochs for VAE intermediate evaluation (default: 20)
phase1_latent_size: dimensionality of latent space (default: 32)
phase1_layer_dims: hidden layer dimensions as integer vector (default: c(512, 128))
phase1_batch_size: batch size for VAE training (default: 256)
phase1_optimizer: optimizer algorithm (default: "Adam")
phase1_lr: learning rate for VAE encoder/decoder (default: 1e-4)
phase1_lr_3: learning rate for additional VAE component (default: 1e-4)
phase1_dropout: dropout rate for VAE layers (default: 0)
phase1_type: VAE layer type, "Dense" or "Normal" (default: "Dense")

Phase 2: Deep Cox training parameters:

phase2_epochs: total epochs for Cox model training (default: 500)
phase2_hidden: number of hidden units in Cox model (default: 128)
phase2_lr: learning rate for Cox model (default: 1e-4)
phase2_dropout: dropout rate for Cox model (default: 0)
phase2_test_size: proportion of data held out for testing (default: 0.2)
phase2_batch_size_bulk: batch size for bulk data in Cox training (default: 256)

Training & risk definition parameters:

train_iterations: number of risk score iteration rounds (default: 5)
train_percentile: percentile threshold for high-risk cell selection (default: 0.95)
train_steepness: steepness parameter for risk score transformation (default: 30)
train_path: directory path for saving intermediate results (default: "./SIDISH_res/")
train_num_workers: number of data loading workers (default: 0)
train_distribution_fit: distribution fitting method, "fitted" or "default" (default: "fitted")

...

Additional arguments passed to the function. Common parameters include:

A named list containing:

scRNA_data: Modified single-cell data object with integrated screening results.