Cross Species Integration and Comparison of single cell RNA sequencing data

This repository is the result of my 3 months internship at the EMBL-EBI. The first aim was to test if the integration tool expiMap was able to perform cross-species integration. Simulations not giving satisfactory results, it suggests that expiMap does not allows cross-species integration. Moving on, the second purpose was to test SATURN, a promising cross-species integration tool. To perform these integrations, I used scRNA-seq data from the primary motor cortex beteween Homo sapiens, Mus musculus and Drosophila, starting with only mouse and human.

Data preprocessing

This concern the repository data. Each file gives an anndata object, .h5ad file, where the data are cleaned if necessary. For the mouse I used the mouse_data_from_cellxgene_yuyao.ipynb file.

Expimap

Anndata input construction

To use expiMap, the first step is to create the input. In expiMap/object_human_mouse there are all the scripts I made to create the final anndata object with the homologous genes from the human and the mouse. The left part of the matrix is the human counts and the right part is the mouse counts.

1. ensembl_query.ipynb: Get the association between the human genes and the mouse genes and create a dataframe with all the homologous genes and their information.
2. one2one_human_mouse.ipynb: Create the o2o part of the matrix and the .var dataframe associated. For each gene, the raw count of each cell from the human are concatenated with the raw count of each cell from the mouse.
3. one2many_human_mouse.ipynb: Create the o2m part of the matrix and the .var dataframe associated. For each gene, the raw count of each cell from the human are concatenated with the counts average of the multiple homologous genes associated from the mouse, for each cell.
4. many2one_human_mouse.ipynb: Create the m2o part of the matrix and the .var dataframe associated. For each gene, the count of each cell from the mouse are concatenated with the counts average of the multiple homologous genes associated from the human, for each cell.
5. many2many_human_mouse.ipynb: Create the m2m part of the matrix and the .var dataframe associated. For each group of homologous genes, the counts average of each cell from the human are concatenated with the counts average of each cell from the mouse.
6. all_homologous_genes.ipynb: Create the final object by concatenating the o2o, o2m, m2o and m2m part as well as their .var dataframe.
7. reduced_matrix.ipynb: Create a reduced version of the final object by keeping 20% of each cell type above a threshold.

Training

The training files are in the expiMap folder, I followed both the basic tutorial and the advanced tutorial. Each type of training has a .ipynb file, to do most of the tutorial and visualise the results, and a .py file which is used to actually performed the training.

• expimap_training.ipynb & expimap.py: first attempt of expimap with the reactome.gmt annotation file provide in the tutorial and with the whole dataset.
• expimap-GABAergic.ipynb & expimap_GABAergic.py: second attempt with only the GABAergic cells and the reactome annotation.
• expimap_advanced_GABAergic_GO.ipynb & expimap_adv_GABAergic_GO_reduced.py: third attemps with only the GABAergic cells using the reduced dataset plus using the Gene Ontology annotation.
• GO_matrix.ipynb: Looking through the GO_BP_human_gene_id_binary_matrix.csv file.

SATURN

The repository SATURN is a clone of the project SATURN. All the original files are still present without modification. All information about SATURN is available in the SATURN/README.md file. I added scripts to use SATURN to perform a cross-species integration with the human and the mouse dataset.

Preprocessing

1. data_preprocessing.ipynb: inspired by the script Vignettes/frog_zebrafish_embryogenesis/dataloader.ipynb, I load each anndata object and apply the modification required to run SATURN including adding the species name, filtering the cells and the genes, putting the gene name as the .var_names. This script also create for the human and mouse 3 other objects each being one specific cell type among GABAergic, Glutamatergic and Non-Neuronal. All the objects are save in the directory Vignettes/frog_zebrafish_embryogenesis/data.
2. gene_names.ipynb: get human and fly information from ensembl, not needed.
3. protein_embeddings/gpe_human_mouse_fly.ipynb: based on the protein_embeddings/Generate Protein Embedding.ipynb script, download the reference proteom for mouse and human and generate the embeddings files. Before generating the embeddings files, gpe_human.sh, gpe_mouse.sh and gpe_fly.sh need to be runned, it requires to download the ESM repository and checkout commit 839c5b82c6cd9e18baa7a88dcbed3bd4b6d48e47.

Running SATURN

All the futur scripts are based on Vignettes/frog_zebrafish_embryogenesis/Train SATURN.ipynb file and are situated in the directory Vignettes/frog_zebrafish_embryogenesis/. The training use the train-saturn.py file where 3 lines where changes from df.append() to pd.concat(), issue due to pandas version.

1. training_human_mouse.ipynb: create the input for training SATURN, use the whole human and mouse datasets and visualise the results. To run the training, use the saturn_training.sh file.
2. training_human_mouse_GABAergic.ipynb: same as training_human_mouse.ipynb but only for GABAergic cells, use the saturn_training_GABAergic.sh file to run the training.
3. training_human_mouse_Glutamatergic.ipynb: same as training_human_mouse.ipynb, only for Glutamatergic cells, use the saturn_training_Glutamatergic.sh file to run the training.
4. training_human_mouse_Non-Neuronal.ipynb: same as training_human_mouse.ipynb, only for Non-Neuronal cells, use the saturn_training_Non-Neuronal.sh file to run the training.
5. results.ipynb: visualise the results of the 4 previous integration, PCA and UMAP plots.
6. training_human_mouse_fly.ipynb: create the input file for training SATURN, use the whole human, mouse and fly objects. To run the training, use the saturn_training_hmf.sh file. Did not manage to make it work for now.

Data availability

Species	Source	Name	Link
Human	Allen brain map	Human M1 10X	https://portal.brain-map.org/atlases-and-data/rnaseq/human-m1-10x
Mouse	CellxGene	10X nuclei v3 Broad	https://cellxgene.cziscience.com/collections/ae1420fe-6630-46ed-8b3d-cc6056a66467
Drosophila	The Fly Cell Atlas	E-MTAB-10519	http://ftp.ebi.ac.uk/pub/databases/microarray/data/atlas/sc_experiments/E-MTAB-10519/

Packages

expiMap

See the tutorials to download the packages required.

SATURN

Use the requirements.txt file and follow the indications present in the SATURN/README.md file.

Packages version

• scanpy==1.9.3
• anndata==0.9.1
• umap==0.5.3
• numpy==1.24.3
• scipy==1.10.1
• pandas==2.0.1
• scikit-learn==1.2.2
• python-igraph==0.10.4
• scarches==0.5.8
• biomart==0.9.2