API

Python API

Processing

Calculate ambient profile

setup_anndata

Calculate ambient profile for relevant features

Training

The core module of scar

model

The scar model

Synthetic_dataset

Generate synthetic datasets (scRNAseq, CITE-seq, scCRISPRseq) with ambient contamination

scrnaseq	Generate synthetic single-cell RNAseq data with ambient contamination
citeseq	Generate synthetic ADT count data for CITE-seq with ambient contamination
cropseq	Generate synthetic sgRNA count data for scCRISPRseq with ambient contamination

Plotting

Plotting functions (under development).

Reporting

Generate denoising reports (under development).

Command Line Interface

scAR (single cell Ambient Remover): denoising drop-based single-cell omics data

usage: scar [-h] [--version] [-ap AMBIENT_PROFILE] [-ft FEATURE_TYPE]
            [-o OUTPUT] [-m COUNT_MODEL] [-sp SPARSITY] [-hl1 HIDDEN_LAYER1]
            [-hl2 HIDDEN_LAYER2] [-ls LATENT_DIM] [-epo EPOCHS] [-d DEVICE]
            [-s SAVE_MODEL] [-batchsize BATCHSIZE]
            [-batchsize_infer BATCHSIZE_INFER] [-adjust ADJUST]
            [-cutoff CUTOFF] [-round2int ROUND2INT] [-clip_to_obs CLIP_TO_OBS]
            [-moi MOI] [-verbose VERBOSE]
            count_matrix [count_matrix ...]

Positional Arguments

count_matrix

The file of raw count matrix, 2D array (cells x genes) or the path of a filtered_feature_bc_matrix.h5

Named Arguments

--version

show program’s version number and exit

-ap, --ambient_profile

The file of empty profile obtained from empty droplets, 1D array

-ft, --feature_type

The feature types, e.g. mRNA, sgRNA, ADT, tag, CMO and ATAC

Default: “mRNA”

-o, --output

Output directory

-m, --count_model

Count model

Default: “binomial”

-sp, --sparsity

The sparsity of expected native signals

Default: 0.9

-hl1, --hidden_layer1

Number of neurons in the first layer

Default: 150

-hl2, --hidden_layer2

Number of neurons in the second layer

Default: 100

-ls, --latent_dim

Dimension of latent space

Default: 15

-epo, --epochs

Training epochs

Default: 800

-d, --device

Device used for training, either ‘auto’, ‘cpu’, or ‘cuda’

Default: “auto”

-s, --save_model

Save the trained model

Default: False

-batchsize, --batchsize

Batch size for training, set a small value upon out of memory error

Default: 64

-batchsize_infer, --batchsize_infer

Batch size for inference, set a small value upon out of memory error

Default: 4096

-adjust, --adjust

Only used for calculating Bayesfactors to improve performance,

‘micro’ – adjust the estimated native counts per cell. Default.

‘global’ – adjust the estimated native counts globally.

False – no adjustment, use the model-returned native counts.

Default: “micro”

-cutoff, --cutoff

Cutoff for Bayesfactors. See [Ly2020]

Default: 3

-round2int, --round2int

Round the counts

Default: “stochastic_rounding”

-clip_to_obs, --clip_to_obs

clip the predicted native counts by observed counts, use it with caution, as it may lead to overestimation of overall noise.

Default: False

-moi, --moi

Multiplicity of Infection. If assigned, it will allow optimized thresholding, which tests a series of cutoffs to find the best one based on distributions of infections under given moi. See [Dixit2016] for details. Under development.

-verbose, --verbose

Whether to print the logging messages

Default: True