Metadata-Version: 2.1
Name: icat-sc
Version: 0.1.3
Summary: Identify cell states across treatments in single-cell RNA sequencing experiments
Home-page: https://github.com/BradhamLab/icat
Author: Dakota Y. Hawkins
Author-email: dyh0110@bu.edu
License: BSD
Description-Content-Type: text/markdown
License-File: LICENSE

# icat
Identifying Cell-states Across Treatments

ICAT is a tool developed to better identify cell states in scRNAseq experiments where perturbations or some other biologic heterogeneity is present, such as gene knock-outs.

The method works by first identifying a set of conrol-defined cell states by performing unsupervised clustering. These identified cell states are then fed into a sparse gene weighting algorithm, Neighborhood Component Feature Selection (NCFS), to highly weight the most predictive genes, while also removing variance from non-explanatory genes. We then transform the data matrix using this weight vector, and perform semi-supervised clustering such that the originally identified control labels remain constant, but cells from experimental conditions

## Installation

ICAT can be installed using `pip` and can be installed using the following command:

`pip install icat-sc`


## How to use

ICAT makes heavy use of the excellent `scanpy` library along with the associated `AnnData` data structure.

An example code block walks through running `icat` on a simulated dataset. The 
final clustering is stored in the `sslouvain` column of the returned `AnnData`
object.

```python
    from icat import simulate
    from icat import models
    import scanpy as sc
    import numpy as np
    data_model = simulate.SingleCellDataset(
        populations=2,
        genes=1000,
        dispersion=np.random.choice([1, 2, 3], 1000)
    )
    controls = data_model.simulate()
    controls.obs['treatment'] = 'control'
    perturbed = simulate.perturb(controls)
    perturbed.obs['treatment'] = 'perturbed'
    adata = controls.concatenate([perturbed])
    sc.pp.log1p(adata)
```
**visualizing dataset**
![](docs/images/raw_input.png)
```python
    # specify model parameters -- see documentation for more information
    model = models.icat(
        ctrl_value="control",
        ncfs_kws={'reg': 1, 'sigma': 3},
        neighbor_kws={'n_neighbors': 15}, 
        cluster_kws={'resolution': 0.75},
    )
    # cluster cells by providing treatment information
    out = model.cluster(adata, adata.obs['treatment'])
    print(out.obs['sslouvain'].unique())
```
**visualizing results**
![](docs/images/icat_output.png)

