chrombert_imputation_cistrome_sc

Generate prediction result (hdf5 format) from ChromBERT when given single cell, region and regulator.

chrombert_imputation_cistrome_sc [OPTIONS] SUPERVISED_FILE --o-h5 H5_PATH --finetune-ckpt CKPT --prompt-kind KIND

Options

supervised_file

Path to the supervised file.

--o-h5

Path of the output HDF5 file. This option is required.

--prompt-kind

Prompt data class. Choose from cistrome or expression. This option is required.

--basedir

Base directory for the required files. Default is set to the value of DEFAULT_BASEDIR.

-g, --genome

Genome version. For example, hg38 or mm10. Only hg38 is supported now. Default is hg38.

--pretrain-ckpt

Path to the pretrain checkpoint. Optional if it could be inferred from other arguments.

-d, --hdf5-file

Path to the HDF5 file that contains the dataset. Optional if it could be inferred from other arguments.

-hr, --high-resolution

Use 200-bp resolution instead of 1-kb resolution. Caution: 200-bp resolution is preparing for the future release of ChromBERT, which is not available yet.

--finetune-ckpt

Path to the finetune checkpoint. Optional.

--prompt-dim-external

Dimension of external data. Use 512 for scGPT and 768 for ChromBERT’s embedding. Default is 512.

--prompt-celltype-cache-file

Path to the cell-type-specific prompt cache file. Optional.

--prompt-regulator-cache-file

Path to the regulator prompt cache file. Optional.

--prompt-regulator-cache-pin-memory

Pin memory for regulator prompt cache for further accelerating. Default is False.

--prompt-regulator-cache-limit

The limit of regulator prompt cached in memory. Be mindful of your memory usage!

--prompt-celltype

The cell-type-specific prompt. For example, dnase:k562 for cistrome prompt and k562 for expression prompt. It can also be provided in the supervised file if the format supports. Optional.

--prompt-regulator

The regulator prompt. Determine the kind of output. For example, ctcf or h3k27ac. It can also be provided in the supervised file if the format supports. Optional.

--batch-size

Batch size. Default is 8.

--num-workers

Number of workers for the dataloader. Default is 8.