Predict query cell types using a trained CellFuse model

Projects query cells into a previously learned embedding space and assigns predicted cell type labels using K-nearest neighbors.

PredictCells(
  dataset_name,
  data_dir,
  test_data_dir,
  test_data,
  model_dir,
  model_date,
  device,
  cluster_column,
  lr,
  margin,
  bs,
  epoch,
  knn_k,
  output_dim,
  dropout_prob,
  activation_function
)

Arguments

dataset_name

Character. Name of training dataset.

data_dir

Character. Path to reference data directory.

test_data_dir

Character. Path to query dataset directory.

test_data

Character. Name of query dataset.

model_dir

Character. Path to saved model directory.

model_date

Character. Timestamp identifying trained model. YYYY-MM-DD

device

Character. "cpu" or "cuda".

cluster_column

Character. Column containing cluster labels.

lr

Numeric. Learning rate.

margin

Numeric. Margin for contrastive loss.

bs

Integer. Batch size.

epoch

Integer. Number of epochs.

knn_k

Integer. Number of neighbors for KNN classification.

output_dim

Integer. Embedding dimension.

dropout_prob

Numeric. Dropout probability.

activation_function

Character. Activation function.

Value

Invisibly returns NULL. Prediction results are written to disk.

Details

This function loads a trained CellFuse model and applies it to a query dataset

The query dataset must:

  • Contain the same markers used during training

  • Have rows corresponding to individual cells

Predicted labels and embeddings are saved to the output directory.

Examples

if (FALSE) { # \dontrun{
PredictCells(
  dataset_name = "CyTOF",
  data_dir = "Reference_Data/",
  test_data_dir = "Query_Data/",
  model_dir = "Predicted_Data/Saved_model"
)
} # }