kmmuleelab/P2Mdb

A curated collection of single-cell and bulk transcriptomic datasets from primary tumors and metastases

P2M.db is a curated collection of single-cell and bulk transcriptomic datasets from primary tumors and metastases, with specimen metadata, cell annotations and analysis results. The data were collected from GEO, ArrayExpress and TCGA.

The collection contains 48 single-cell datasets (3,648,483 cells) and 86 bulk datasets (12,896 expression profiles), covering 49 cancer types and subtypes, 28 primary tissues and 25 metastatic tissues.

Website · Browse the files

Figure 1. Overview of the transcriptomic collection, cancer coverage, deposited records, specimen roles and data access.

Single-cell quality control · Bulk quality control · Figure legends

The complete search queries and accession-level GEO screening table, including screening decisions and exclusion counts by reason, are available for download.

Data files

Datasets are organized by their original study accession under scRNA-seq/ and bulk/.

Expression data are provided in two formats: RDS files and CSV files packaged in TAR.GZ archives. Both formats include the expression data and corresponding cell or sample metadata.

File or directory	Contents
.rds	Expression data and metadata; single-cell objects also contain available cell annotations
<accession>_expression.tar.gz	TAR.GZ archive containing expression.csv and metadata.csv
samples.csv	Sample identifiers, specimen classifications and associated metadata
dataset.csv	Study information, data source, annotation methods and processing notes
qc.csv	Quality control measurements
processing_parameters.csv	One function per row in two columns: function and parameters
analysis/	Available differential expression, enrichment and other analysis results

Each parameter table has one function per row, with its arguments listed as name = value pairs in the parameters column. PCA/Harmony contexts retain separate labels. Functions marked (workflow reference) give settings from the original pipeline, rather than recovered execution records for that accession. Multiple source descriptions of a parameter are retained as a list. The format can be converted with code/parameters.py; no expression values or analysis results are changed by this formatting.

Each expression archive contains expression.csv (features in rows; cells or samples in columns) and metadata.csv. Metadata rows follow the expression column order and retain specimen labels and available cell annotations. Single-cell exports use the stored RNA counts layer when available, otherwise the stored expression data layer; expression_layer identifies the selection. Bulk expression values retain their recorded scale. Studies with multiple bulk matrices contain a separate CSV pair for each platform. Missing values are written as NA.

Available analyses differ among datasets. analysis/comparisons.csv identifies the specimen groups, metastatic sites and annotation methods used in each comparison.

At the repository root, sources.csv records study sources and the evidence for specimen labels and annotations. Collection statistics are in statistics/.

Read the data

Read RDS files directly in R, or extract the TAR.GZ archive before reading its CSV files:

object <- readRDS("path/to/file.rds")
metadata <- read.csv("path/to/metadata.csv", check.names = FALSE)

For examples that select specimens and handle bulk objects with multiple matrices, see code/read_data.R and code/README.md. Analysis scripts and their documentation are in code/. The original P2M.Rcode analysis scripts are mirrored in code/analysis/ with standardized filenames and unchanged contents.

Use of the data

Check the specimen labels and comparison definitions before analysis. Metastatic tissue samples and TCGA clinical M0/M1 groups represent different comparisons. Some datasets also contain normal, fluid or unclassified specimens. Sample counts are not necessarily counts of independent patients.

Some historical analysis tables lack complete input sample records; the dataset-specific limitations are recorded in sources.csv and dataset.csv. The 20 TCGA expression releases contain TPM values; the bulk analysis workflows use log2(TPM + 1). QC summaries describe the released matrices or the recorded filtering stages. Unavailable measurements and annotations are represented by NA or empty fields.

Data use and citation

The curated metadata, QC summaries and analysis results generated for P2M.db are licensed under CC BY 4.0. Original expression data and source annotations retain their source terms; code and reference resources retain their own licenses. See LICENSE for the scope.

Cite this repository and the original studies listed in sources.csv, and record the repository revision used in your analysis.

TCGA survival analyses

Overall-survival inputs, all tested gene models and model diagnostics for 18 TCGA cohorts are provided in each bulk/TCGA_*/survival/ directory; the combined gene results are in analysis/survival.csv.gz. Death and last-follow-up times are retained, with one tumor profile per patient. Analyses use clinical M0/M1 groups, which are distinct from specimen tissue origin. Nominal and BH-adjusted Cox P values and proportional-hazards diagnostics are reported separately. Website Kaplan-Meier curves use a median expression split and a log-rank test.

The current survival workflow is in code/validation/survival/. GEO survival files retain their source-specific endpoints and historical subsets; the TCGA update does not assert that those GEO analyses were reproduced.