Alveolar macrophage signatures are associated with T-cell organization and immunotherapy outcomes in lung cancer

Dataset for: Alveolar macrophage states, T-cell context and immunotherapy outcomes in lung adenocarcinoma This repository contains the analysis code used to generate the results and figures of the manuscript submitted to Nature Communications. It covers the following analyses: • Single-cell atlas. Processing and integration of 1.25 million single cells from Human Tumor Atlas Network (HTAN) LUAD cohorts and the LuCA extended atlas. This includes quality control, doublet removal, CellTypist annotation, separate clustering of LUAD and normal-lung cells, and identification of alveolar macrophage (AM) states AM-1 to AM-4. • AM state characterization. Expression programs (NMF), module scores, PROGENy pathway and CytoSig cytokine inference, cell–cell communication, contact-molecule comparisons and scPagwas genetic associations. • Spatial analyses. Visium CytAssist deconvolution and spatial programs across LUAD progression stages, and HALO-based multiplex immunofluorescence analyses of in-house treatment-naive and neoadjuvant chemo-immunotherapy cohorts (cell densities, nearest-neighbor distances and association with pathological response). • TWAS. Transcriptome-wide association study in the OTTERS framework using GTEx lung eQTL weights and ACAT, and construction of the TWAS score. • Bulk-cohort and clinical analyses. Construction of the AM score (contrast of the AM-2 and AM-4 signatures), correlation analyses across eight LUAD/NSCLC cohorts, and exploratory trial-stratified Cox models of the relative progression-free survival effect of atezolizumab versus docetaxel in the OAK and POPLAR trials. • Figure scripts. Scripts for all main and supplementary figures. Data. No raw or patient-level data are included. Public data must be obtained from the original sources, as listed in the manuscript’s Data availability statement: • authorized-access data: HTAN (dbGaP phs002371), and OAK and POPLAR (EGA EGAD00001008549 and EGAD00001008548); • open data: CELLxGENE, GEO, Zenodo, GDC, the GWAS Catalog and the GTEx Portal. In-house mIF data are available from the corresponding authors on reasonable request. Software environment. Analyses were run in Python (scanpy, scikit-learn, CellTypist) and R (Seurat, survival). Package versions are given in the manuscript Methods and in the repository’s environment files. Citation. If you use this code, please cite the associated article (citation to be updated upon publication).