An interpretable and adaptive autoencoder for efficient tissue deconvolution

An interpretable and adaptive autoencoder for efficient tissue deconvolution. 

This paper is available in bioRxiv (https://arxiv.org/pdf/2311.11991) and is currently under review. The github repository is: (https://github.com/ML4BM-Lab/Sweetwater/tree/main)

Here we provide the 4 datasets used along the Sweetwater paper.  In order to reproduce the results and run Sweetwater with every dataset: 

Call load_X.py, being X the dataset/subdataset used. e.g. for the PBMC GS, load_pbmc_gs_data.py

This will return 4 elements: scRNA-seq, bulkRNA-seq, common_genes, bulkrna_props

scRNA-seq: scRNA-seq reference expression matrix.

bulkRNA-seq: bulkRNA-seq matrix to be deconvolved.

common_genes: genes that both matrix have in common, hence defining the input size of the model.

bulkrna_props: proportions of the bulkrna-seq matrix to be deconvolved.

run python3 src/main.py with the path to both the scRNA-seq and bulkRNA-seq path. (see github readme)

Get the deconvolved proportions. Afterwards, you can evaluate the performance using bulkrna_props.