Code repository for the paper "Lipidomics Identifies HFpEF Phenogroups and a High-Risk Metabolic Signature" - The BECAME-HF project

Clustering analysis

The script clustering.R performs unsupervised clustering of patients with heart failure with preserved ejection fraction (HFpEF) using lipidomics data and evaluates associated clinical outcomes through survival analysis.

The workflow integrates:

• Principal Component Analysis (PCA)
• Hierarchical Clustering on Principal Components (HCPC)
• Kaplan-Meier survival analysis

Running the script requires the input file input_data_clustering.xlsx available here.

Dependencies

install.packages(c(
  "ggplot2",
  "readxl",
  "dplyr",
  "FactoMineR",
  "factoextra",
  "survival",
  "survminer"
))

For the consensus clustering analysis using ClustOmics, see package here.

Cross-Cohort Comparison and Lipid Signature

Analysis pipeline comparing lipidomic profiles across the Belgian (BECAME-HF1) and Canadian (BECAME-HF2) HFpEF cohorts. The second part of the analysis aims at identifying a minimal lipid signature distinguishing the B1 patient cluster from other subjects in the BECAME-HF1 cohort.

Pipeline

Running the pipeline requires to download the Supplemental data on the Mendeley repository (doi:10.17632/rnhdrhsxsz.2). All steps can be run individually but they all use the processed data from 0_CleanDataset.R, which would need to be run at least once. 3_CorrelationGraph.R also needs 2_MinimalSignatureLasso.R to be run before because it uses the lipids identified for esthetic of the network graph. 4_FigureCreation.R requires all scripts to have run once, since it gathers the results and creates the final figure. The recommended way is to run them sequentially from the bash script:

bash runall.sh

Step	Script	Description
0	0_CleanDataset.R	Reads raw lipidomic data from both cohorts, merges them, removes batch effects with limma, and exports a normalized expression matrix with metadata.
1	1_ClusterClassificationRF.R	Trains a Random Forest on BECAME-HF1 patient clusters (B1/B2/B3) and applies it to reassign BECAME-HF2 patients into the same cluster space.
2	2_MinimalSignatureLasso.R	Runs repeated LASSO regressions to identify lipids that most frequently distinguish cluster B1 from the rest of BECAME-HF1 samples, producing a ranked predictor list.
3	3_CorrelationGraph.R	Builds a lipid–lipid correlation network from the BECAME-HF1 HFpEF patients, highlighting the frequent LASSO predictors within the graph.
4	4_FigureCreation.R	Generates manuscript figures: cluster heatmaps, PCA plots with RF-predicted clusters, ridge-regression probability scores, and feature boxplots.

Configuration

All parameters (paths, thresholds, plot dimensions, colours) are set in config/config.yaml.

Dependencies

R 4.3.2 with the following packages:

Package	Version		Package	Version
car	3.1-3		igraph	2.0.2
caret	6.0-94		limma	3.58.1
cowplot	1.1.3		pheatmap	1.0.12
data.table	1.15.0		plyr	1.8.9
edgeR	4.0.16		pROC	1.18.5
ggfortify	0.4.16		randomForest	4.7-1.2
ggplotify	0.1.2		RColorBrewer	1.1-3
ggpubr	0.6.0		readxl	1.4.3
ggrepel	0.9.5		reshape2	1.4.4
ggridges	0.5.6		tidyverse	2.0.0
glmnet	4.1-8		yaml	2.3.8