TPHP

This repository contains the core analysis code for the TPHP project, which supports large-scale, spatially resolved human proteome profiling. The dataset quantifies >13,000 proteins across 2856 samples spanning 58 major tissue types (251 tissue subtypes) and 25 cancer types using DIA-MS.

The study is described in the bioRxiv (2025) preprint.

The code performs per-cancer, per-protein tumor versus paired non-tumor comparisons using linear mixed-effects regression, enabling systematic analysis of oncogenic proteome changes across tissues. Results are exported as RDS objects for downstream statistical analysis, visualization and integration with the public proteome database.

Repository layout

• tumor_dysregulation_analysis.R — main analysis script
• data/tumor_compare_data.parquet — input table (paired tumor/non-tumor)
• output/compare_report_output.rds — main result
• output/package_versions.txt — R and package versions used

1. System requirements

1.1 Operating systems

The workflow is expected to run on every common desktop/server platforms that support R. The test environment is on Windows 11 x64 system.

1.2 Software dependencies

Required

• R (recommended R 4.0+)
• CRAN packages
  • tidyverse
  • arrow
  • lme4
  • lmerTest
  • plyr

Exact tested version recorded in output/package_versions.txt

1.3 Hardware

• CPU-only execution (no GPU required)
• Recommended RAM: ≥8 GB (larger datasets may require more)
• No required non-standard hardware

2. Installation guide

2.1 Get the code

git clone <REPO_URL>
cd <REPO_DIR>

2.2 Install R

Install R for the operating system from CRAN.

Optionally, install packages manually:

install.packages(c("tidyverse","arrow","lmerTest","lme4"))

This project is script-based, and no software installation step is required beyond installing dependencies.

3. Demo

Quick start

From the repository root:

Rscript tumor_dysregulation_analysis.R

3.1 Input files

Default input path:

• data/tumor_compare_data.parquet

3.2 Expected output files

After a successful run, the script writes:

• output/compare_report_output.rds

  An R list DEA with:

  • DEA$Diff.report: full per-(cancer, protein) model results
  • DEA$Diff.report.filter: filtered results using Hedges'g >= 0.5 and p_adj_BH < 0.05

• output/package_versions.txt R version and package versions used.

3.3 Expected runtime

Runtime scales primarily with the number of model fits, i.e., cancer types × proteins × samples.

As a rough guide, on a standard desktop CPU (8–16 threads, 16–32 GB RAM), throughput is ~25 model fits per second, corresponding to < 10 minutes per cancer type under typical settings.

4. Instructions

4.1 Input data format

The script expects a Parquet table with metadata columns:

• patient_ID (string)
• sample_type (must include NT for non-tumor and T for tumor)
• cancer_abbr (string)
• cancer_subtype (string; may be constant within a subset)
• Gender (categorical)
• Age (integer-like)
• Dataset (categorical)

All remaining columns are treated as numeric features (proteins).

4.2 What the script does

For each cancer_abbr and each protein, the script fits a mixed-effects model:

• value ~ 1 + sample_type + (optional covariates) + (1 | patient_ID)

Optional covariates among {cancer_subtype, Gender, Age_c, Dataset} are included only when estimable within the cancer/protein subset.

The reported tumor effect is the fixed-effect coefficient for sample_typeT.

4.3 Output columns

Each row in DEA$Diff.report corresponds to one (cancer, protein) fit and includes:

• effect, se, t
• df(degree of freedom)
• p, p_adj_BH (p value)
• sigma (residual SD)
• es_adj = effect / sigma (standardized effect)
• g_adj = J(df) * es_adj (small-sample adjusted standardized effect)
• formula (the exact model formula used)
• is_singular, re_var_patient (fit diagnostics)