worm-glia-atlas

This is a repository with tutorials and reproducibility notebooks for the worm glia scRNA-seq atlas available at https://wormglia.org.

Environment Setup

---

To run the tutorial, clone this repository and follow the environment set up below.

git clone https:/settylab/worm-glia-atlas.git

Option I: Creating a conda environment using the provided yaml file:

envName=worm-glia-atlas

conda env create -n "$envName" --file envs/environment.yaml

conda activate "$envName"

Option II: Create an environment and install the required packages using pip:

envName=<your-environment-name>

conda create -n "$envName" python=3.8.10 pip=21.1.3

conda activate "$envName"

pip install -r envs/requirements.txt

Important Dependencies

The following packages need to be installed for running the notebooks, which can also be installed by following the environment setup above:

1. scanpy: https://scanpy.readthedocs.io/en/stable/installation.html
2. sklearn: https://scikit-learn.org/stable/install.html
3. plotly: https://plotly.com/python/getting-started/
4. tqdm : https:/tqdm/tqdm#installation

Pairwise Differential Expression Analysis

---

The tutorial notebook for pairwise differential expression analysis is available here. Pairwise differential analysis performs pairwise differential analysis to identify cluster enriched genes rather than one-vs-all approach.

Inputs

The input is an anndata object with normalized, log-transformed data and an obs variable containing information about clusters for pairwise comparison.

1. anndata.obs[LEIDEN_NAME]: Where LEIDEN_NAME is the name of the column in anndata.obs field containing the groups to be used for the anlaysis (leiden clusters in this analysis).

Outputs

The anndata object is updated with the following information

1. anndata.varm['pairwise_cluster_count']: Gene X Cluster matrix indicating how many comparisons the gene is differential in.
2. anndata.varm['cluster_means']: Gene X Cluster matrix of mean expression of gene per cluster.
3. HTML files of the pairwise analyses results can saved using the plot_pairwise_results() function by specifying a path to the save parameter.

Feature Ranking Analysis

---

The tutorial notebooks for Sheath/Socket marker analysis is here & for Pan-Glia marker analysis is available here. A logistic regression model is trained and employed for binary classification of cells using gene expression. Sheath/Socket notebook describes identification of markers for both classes where Pan-glia notebook describes identification of markers only for the positive class.

Subsequently, a ranking of the learned features within the model is then performed with the objective being to rank features that are highly informative and correlated with the specified target classes or cell type.

These analyses can be readily extended to other datasets by providing the appropriate inputs, as outlined below.

Inputs

The key inputs for this analyses is the anndata object with normalized & log-transformed counts, imputed gene expression values as well as the following anndata attribute fields below:

• anndata.obs[CLASS_LABELS]: Where CLASS_LABELS is the name of the column in anndata.obs containing the ground truth labels for each cells in the anndata object.
• anndata.obs[CLUSTER_LABELS]: Where CLUSTER_LABELS is the name of the column in anndata.obs containing the cluster labels for each cells in the anndata object.
• anndata.var[USE_GENES]: Where USE_GENES is the name of the column in anndata.var containing boolean values that specifies whether a gene is to be used for analysis or ignored (default is highly_variable genes columns).
• anndata.layers[USE_LAYER]: Where USE_LAYER is a key in anndata.layers dictionary corresponding to the imputed Cell X Gene count matrix. If not specified, will use the normalized and log-transformed counts as values for the constructed feature matrix & feature ranking analysis.

Outputs

The output of the analysis is a trained logistic regression model and an updated anndata object as follows:

• anndata.uns['<target_class>_marker_results']: A dictionary object is stored in anndata.uns, containing results of the feature ranking analysis specific to a designated target class.

• anndata.uns['<target_class>_probEst_Summary']: A DataFrame object is stored in anndata.uns, containing the mean probability estimates for each cluster belonging to the specified target class.

• anndata.uns['<target_class>_AUROCC_Summary']: A DataFrame object is stored in anndata.uns, containing summary information about the AUROCC (Area Under the Receiver Operating Characteristic Curve) scores for each cluster belonging to the specified target class.

• anndata.obs['data_splits']: A new column in anndata.obs containing labels indicating whether each cell belongs to the training, validation, or test dataset after the feature matrix and target vector are split accordingly.

• anndata.uns['model_selection_metrics']: A DataFrame object is stored in anndata.uns, containing mean accuracy scores of trained regularized models on the training, validation, and test datasets.

Citations

---

Worm glia atlas manuscript is available on bioRxiv. Please cite our paper if you use these tutorials for your analyses:

@article {Purice2023.03.21.533668,
	author = {Maria D. Purice and Elgene J.A. Quitevis and R. Sean Manning and Liza J. Severs and Nina-Tuyen Tran and Violet Sorrentino and Manu Setty and Aakanksha Singhvi},
	title = {Molecular heterogeneity of C. elegans glia across sexes},
	elocation-id = {2023.03.21.533668},
	year = {2023},
	doi = {10.1101/2023.03.21.533668},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2023/03/24/2023.03.21.533668},
	eprint = {https://www.biorxiv.org/content/early/2023/03/24/2023.03.21.533668.full.pdf},
	journal = {bioRxiv}
}