How gauravj49/scrnaseq works
Main Components
• Data Ingestion and Raw Data Input: This would include all the raw input (for example, outputs from CellRanger, transcriptome files, etc.). Many scripts (with “cellLine”, “StomachBaseline”, “custom_transcriptome”, “mitGFP”, etc. in their names) indicate differing experimental protocols and raw data sources.
• Preprocessing Pipeline: Group together the preprocessing scripts (e.g. “01.x_xx_preprocessing_usage.sh” from multiple pilots) as an initial stage. This stage might include quality control and filtering.
• Normalization and Imputation: The R scripts (such as “R_norm.R” and others dealing with imputation/normalization) are dedicated to processing the raw counts.
• Clustering, Tissue and Tumor Identification: Several scripts (including “clustering_usage.sh,” “tissue_preprocessing_usage.sh” and “tumor_identification_usage.sh”) indicate stages where the data are clustered and assigned to sample groups.
• Annotation: Components for cell-type or sample annotation appear (e.g. “R_annotate_cells_using_singleR.R” and “R_interactive_singleR_annotation.R”), which call external annotation libraries (SingleR).
• Integration and Cross-Species Mapping: The presence of “R_convert_human_to_mouse_genes.R” suggests a layer of integrating data between human/mouse systems.
• Visualization and Reporting: The “simple-example-chart-apps-master” folder (with multiple Dash applications – 3d plots, bar plots, scatter plots, etc.) seems to be responsible for displaying the results interactively.
Representing the Data Flow
• Next, indicate the flow from preprocessing into normalization/imputation and then onward into clustering.
• From the clustering stage, have branches that go to tissue/tumor identification and cell annotation.
• Annotation and integration modules (for example, cross-species mapping) to the visualization layer.
Important Interactions and Technologies
• Mention integration with external tools: “cellRanger” (via usage scripts), “scanorama”, “TrVAE” (a deep learning tool for integration) and perhaps “seurat” (as implied by related CSV outputs).
• For visualization, note that the Dash apps (built on Python’s Plotly Dash) provide interactive reporting.
Architectural Patterns/Design Principles
• The project follows a pipeline architecture. Each stage (ingestion, preprocessing, analytics, visualization) is a discrete component with defined inputs and outputs.
• There is an emphasis on reproducibility with versioned backups and “usage_extra” directories.
• Modular design: separate directories for different tasks (R scripts vs. Python modules) and clear naming conventions that denote pipeline steps.