Sex-Specific Subphenotyping of HFpEF

Multi-omic (proteomics + clinical phenomics) patient stratification in heart failure with preserved ejection fraction (HFpEF), analyzed separately for women and men, with external validation in an independent cohort.

Pipeline

1. Data integration — proteomic and clinical/phenomic feature panels are imputed, filtered (near-zero-variance and highly correlated features removed), and combined via Similarity Network Fusion (SNF).
2. Clustering — spectral clustering on the fused network defines candidate patient subgroups (2–10 clusters evaluated).
3. Survival validation — Cox proportional hazards models and Kaplan-Meier curves test whether the discovered clusters differ in worsening-heart-failure risk.
4. Baseline characterization — clinical/lab tables comparing clusters.
5. Out-of-cohort validation — an XGBoost classifier is trained on the discovery-cohort cluster labels and used to project cluster membership onto an independent validation cohort, whose survival outcomes are then checked against the same clustering.
6. Transcriptomics — RNA-seq counts for the discovered clusters are tested for differential expression, followed by KEGG/Reactome gene set enrichment analysis (GSEA).

Steps 1–5 are run once for the female subgroup and once for the male subgroup.

Data availability

This repository contains code only. The analysis uses patient-level clinical, proteomic, and RNA-seq data from the MYOVASC and Gutenberg Health Study (GHS) cohorts, which are not publicly available due to participant privacy. The data-loading calls near the top of the script (read_SQL_MyoVasc_BL, read_SQL_A6_data, etc.) depend on internal database access and institution-specific helper scripts and will not run outside that environment.

This script is shared to document the analysis methodology for reproducibility/review purposes, not as a runnable end-to-end pipeline.

Configuration

All machine- and institution-specific paths have been factored out into environment variables at the top of the script, rather than hardcoded:

Variable	Purpose
PROJECT_DIR	Working directory for the project
MYOVASC_SQL_DIR	Location of MYOVASC data-loading scripts
GHS_SQL_DIR	Location of GHS data-loading scripts
SHARED_FUNCTIONS_DIR	Location of shared helper functions (e.g. baseline13.r)
PROJECT_DATA_DIR	Location of local data extracts (defaults to <PROJECT_DIR>/data)
PROJECT_OUTPUT_DIR	Where results (RDS/RData) are written (defaults to <PROJECT_DIR>/output)
DB_USER	Database username for internal SQL-loading functions

Set these via a local .Renviron file (which you should not commit) before running, e.g.:

PROJECT_DIR=/path/to/project
MYOVASC_SQL_DIR=/path/to/myovasc/r_sql
GHS_SQL_DIR=/path/to/ghs/r_sql
SHARED_FUNCTIONS_DIR=/path/to/shared/r_functions
DB_USER=your_username

Requirements

R packages: SNFtool, survival, survminer, survcomp, flexclust, ggplot2, Rtsne, rstatix, stringr, janitor, ggsurvfit, readxl, caret, Hmisc, xgboost, pROC, PRROC, dplyr, clusterProfiler, org.Hs.eg.db, ReactomePA, enrichplot.