Enhanced scRNA-seq Pipeline for Pediatric Leukemia

LeukoMap is a modern re-analysis of the Caron et al. (2020) pediatric leukemia single-cell RNA-seq dataset, from the UConn scRNA class. This project aims to extend the original analysis with deeper embeddings, clinical references, richer annotations, and functional drug mapping.

• Understand intra-individual transcriptional heterogeneity in pediatric ALL
• Link developmental state to druggable biology
• Build a reusable, modular scRNA-seq pipeline for future leukemia engineers

analyze(scRNA_seq_data, healthy_reference) -> annotated_clusters + druggable_targets

-> load_data(data_dir) -> AnnData
-> preprocess(AnnData) -> AnnData
-> train_scvi(AnnData) -> scVI_model
-> embed(scVI_model, AnnData) -> latent_space
-> load_reference(reference_path) -> AnnData
-> align_with_reference(latent_space, reference_AnnData) -> aligned_AnnData
-> auto_annotate(aligned_AnnData) -> cell_type_labels
-> compare_annotations(auto_labels, manual_labels) -> annotation_metrics
-> launch_cellxgene(aligned_AnnData) -> interactive_dashboard
-> run_pseudotime(aligned_AnnData) -> pseudotime_scores
-> differential_expression(aligned_AnnData) -> DEG_table
-> run_gsea(DEG_table) -> enriched_pathways
-> query_lincs(DEG_table) -> druggable_targets
-> export_outputs(AnnData, DEG_table, pathways, druggable_targets) -> annotated_clusters + druggable_targets files (CSV, JSON, PDF)
-> package_pipeline(configs) -> Nextflow_workflow

1. Clone the repository:

git clone https://github.com/bmwoolf/leukomap.git
cd leukomap

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

Pretrained scVI model weights for LeukoMap are available on Hugging Face Hub: https://huggingface.co/bmwoolf/leukomap-scvi

You can load the model directly in your code as follows:

from scvi.model import SCVI
import scanpy as sc

# Load your AnnData object
adata = sc.read_h5ad("your_data.h5ad")

# Load the pretrained model from Hugging Face
model = SCVI.load("bmwoolf/leukomap-scvi", adata)

# Get latent representation
latent = model.get_latent_representation()

Please see ./adata/adata_example.js and ./adata/adata_example.py for example outlines of how you would represent real world cell and gene annotations in memory. Pretty cool.

1. Clone the repository:

git clone https://github.com/bmwoolf/leukomap.git
cd leukomap

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

• GEO: GSE132509 — Caron et al. (2020)
• Original paper
• Class source