Data Descriptor: Raw Single-Cell Transcriptomic Data and Full Implementation Code for Double Machine Learning Causal Inference Benchmarking

This data descriptor documents all raw, preprocessed, analytical datasets and complete open-source codebase supporting the study Comparison of nuisance function construction strategies for double machine learning causal inference in single-cell transcriptomics: shared unsupervised deep learning does not require cross-fitting. All resources enable full reproducibility of the 2×3 factorial experiment comparing three nuisance estimation pipelines (S1 direct linear regression, S2 shared autoencoder representation, S3 dual independent supervised neural networks) under two fitting paradigms (in-sample training vs. 5-fold cross-fitting) for gene-disease causal effect estimation in systemic lupus erythematosus (SLE) memory B cells.

The dataset, preprocessing workflows, deep learning architectures, DML orthogonal score calculation, evaluation pipelines, and visualization scripts are fully archived for transparent replication and extension to other single-cell causal inference tasks.