DeSide: A unified deep learning approach for cellular deconvolution of bulk tumors based on scRNA-seq data

DeSide workflow This is the complete workflow and results of the DeSide project, which includes the following components: main_workflow.py: The main workflow of the DeSide project. TCGA_dataset_preprocessing.py: The script used for TCGA dataset prepocessing. datasets/*: The datasets used in the DeSide project, including the TCGA dataset, the synthesized bulk RNA-seq datasets, and the single cell RNA-seq datasets. model_evaluation/*: The predicted cell proportions of different methods for performance evaluation (used for debugging). results/*: The results of the DeSide project, including the predicted cell proportions by DeSide with different datasets or parameters. plot_fig_update/*: The scripts for plotting the figures in the DeSide project. tables_model_evaluation/*: The predicted cell proportions of different methods. The file structure of the project DeSide_workflow├── LICENSE├── README.md├── TCGA_dataset_preprocessing.py├── correct_by_total_rna.py├── datasets│ ├── Independent_Test_Set│ ├── TCGA│ ├── cancer_purity│ ├── gene_set│ ├── generated_sc_dataset_12ds_n_base100_all_subtypes│ ├── simulated_bulk_cell_dataset_subtypes_all_range│ └── single_cell├── main_workflow.py├── model_evaluation│ ├── Distribution_of_cell_types.ipynb│ ├── Fig.3a_compare_different_algos.py│ ├── Fig.3c_predict_cell_prop_other_test_set.py│ ├── cibersortx_preprocessing.py│ ├── dis_cell_types_cibersort_kassandra.png│ └── predicted_cell_fractions├── plot_fig_update│ ├── Attached_figures-Copy1.ipynb│ ├── Attached_figures.ipynb│ ├── Attached_figures_previous_vs_new_model.ipynb│ ├── Fig.1 - PCA Plot.ipynb│ ├── Fig.1.ipynb│ ├── Fig.2.ipynb│ ├── Fig.3.ipynb│ ├── attached_figs│ ├── data_in_figures│ ├── fig1│ ├── fig2│ ├── fig3│ ├── fig4│ ├── loo_mmd_distances.csv│ ├── merged_sc_ds_sample_count.csv│ └── sup_figures├── results│ ├── whole_workflow_20230216_7ds_new_hp_without_caner_cell│ ├── whole_workflow_20231110_12ds_subtypes│ ├── whole_workflow_20240119_12ds_subtypes│ └── whole_workflow_20240120_12ds_subtypes└── tables_model_evaluation └── predicted_cell_fractions 26 directories, 19 files Dataset synthesis We have also included the workflow documents (.pdf) and demonstration code (.ipynb) for synthesizing the D0, D1, and D2 datasets within the ‘Synthetic_Bulk_Tumor_RNA-seq_Data.zip’ file. This archive contains the following files: Synthetic_Bulk_Tumor_RNA-seq_Data ├── Synthesis Framework for D0 datasets.pdf ├── Synthesis Framework for D1 datasets.pdf ├── Synthesis Framework for D2 datasets.pdf ├── example_for_synthesize_D0_dataset.ipynb ├── example_for_synthesize_D1_dataset.ipynb └── example_for_synthesize_D2_dataset.ipynb 1 directory, 6 files After configuring the environment for DeSide and downloading the necessary input files, as outlined in the corresponding pdf files, users can directly execute the script files to synthesize D0, D1, and D2 datasets using the default parameters. Additional Resources and Documentation The GitHub repository for DeSide: https://github.com/OnlyBelter/DeSide The complete documentation for DeSide can be found at: https://deside.readthedocs.io/en/latest/index.html More examples of DeSide usage can be found at: https://github.com/OnlyBelter/DeSide_mini_example All the datasets, including the collected scRNA-seq datasets and synthesized bulk RNA-seq datasets, are listed here: https://deside.readthedocs.io/en/latest/datasets.html