Aggregate recovery models across runs

Aggregate multiple recovery model weight matrices by identifying consistently selected marker genes across models and averaging their weights.

Usage

AggregateRecoverModels(
  model_list,
  delta.threshold = 0.01,
  min.model.fraction = 0.5
)

Arguments

model_list

A list of recovery model objects. Each element should contain a named list of weight matrices indexed by cell type, derived from `NMFGenerateWList`.

delta.threshold

Minimum delta weights required for a gene to be considered associated with a spatial ecotype. Default is `0.01`.

min.model.fraction

Minimum fraction of models in which a gene must be selected to be retained. Default is `0.5`.

Value

A named list of aggregated weight matrices, one per cell type.

Details

This function is designed for aggregating recovery models generated from repeated training runs or cross-validation folds. For each cell type, genes are retained if they are consistently selected across models based on:

• a minimum delta threshold required for a gene to be selected

• a minimum fraction of models in which the gene is selected

Positive and negative feature pairs (`__pos` and `__neg`) are retained for all selected genes. Final weights are computed as the mean across non-missing values.

Examples

if (FALSE) { # \dontrun{
Ws_list = lapply(1:30, function(ii){
  Ws = NMFGenerateWList(scdata, scmeta,
                        CellType = "CellType",
                        SE = "SE", Sample = "Sample",
                        nfeature = 300,
                        nfeature.per.se = 50,
                        ncores = 8)
})
aggregated_models <- AggregateRecoverModels(
  model_list = Ws_list,
  loading_margin = 0.01,
  min_model_fraction = 0.5
)
} # }