This repository contains all source code and processed data required to reproduce the figures and supplementary results presented in:
LipidTrend: A Structure-Aware Framework for Detecting Continuous Trends in Lipidomic Remodeling
LipidTrend is a structure-aware statistical framework for detecting continuous lipidomic remodeling in chain-length × double-bond structural space.
This repository provides:
- All R scripts used to generate main and supplementary figures
- Processed input data (as described in the manuscript)
- A fully locked computational environment via
renv - A fixed random seed, so every figure and table is bit-for-bit reproducible
No new datasets were generated in this study. All lipidomics datasets were obtained from previously published studies as detailed in the manuscript.
All analyses were performed under:
- R 4.5.2
- LipidTrend 1.3.1 (shipped as a source tarball in
renv/cellar/) - Bioconductor 3.22
- All other package versions pinned in
renv.lock
To reproduce the computational environment:
install.packages("renv")
renv::restore()renv::restore() installs every package at the exact version recorded in
renv.lock, including LipidTrend 1.3.1 from renv/cellar/. Start R from the
repository root so that .Rprofile activates renv automatically.
renv pins R package versions, but not the BLAS library that R uses for
matrix arithmetic. Different BLAS implementations sum the same matrix product
in different orders and disagree in the last few bits of the result.
These results were produced with R's own reference (netlib) BLAS, which is the recommended configuration for reproducing them. Check which BLAS your R is using:
extSoftVersion()[["BLAS"]]A path ending in libRblas is the reference BLAS, and is the default for the
official R builds on macOS and Windows — nothing further to do. Many Linux
distributions instead route R through an optimised library. If the value above
names FlexiBLAS, switch to the netlib backend at the start of your session:
flexiblas::flexiblas_switch(flexiblas::flexiblas_load_backend("NETLIB"))Otherwise, select the reference build system-wide before starting R — on
Debian and Ubuntu, update-alternatives --config libblas.so.3 offers it.
renv.lock is platform-independent; the renv/library/ directory is not, and
is deliberately excluded from version control. Each machine builds its own
library from the lockfile.
On Linux, most CRAN packages will be compiled from source unless you point
renv at a binary repository. Posit Package Manager serves Linux binaries and
is much faster:
options(repos = c(CRAN = "https://packagemanager.posit.co/cran/__linux__/jammy/latest"))
renv::restore()Substitute your distribution's codename for jammy. LipidTrend itself is a
pure-R package with no compiled code, so it installs from the bundled tarball
on any platform.
To reproduce every figure and table:
Rscript run_all.RThis runs all 27 analysis scripts, each in its own R process, writing results
to results/, per-script output to logs/, and a status table to
logs/run_summary.tsv.
To reproduce a single figure, run its script from the repository root:
source("scripts/<dataset>/<comparison>/<lipid class>/<script>.R")All scripts:
- Load data from
data/ - Set a fixed random seed before each permutation test
- Perform structure-aware smoothing and permutation testing
- Apply BH-FDR correction
- Save figures and tables to
results/
Every script sets RNGkind() explicitly and calls set.seed(1234)
immediately before each analyzeLipidRegion() call. Re-running any script
reproduces its CSV output byte-for-byte. PDF outputs differ only in the
/CreationDate and /ModDate metadata that R's PDF device embeds at write
time; their rendered content is identical.
See REPRODUCIBILITY.md for the exact environment, the verification that was carried out, and the changes made relative to the previously published scripts.
Each analysis follows the workflow described in the manuscript:
- Species-level statistical testing.
- Signed log-transformed regional scoring.
- Gaussian kernel smoothing in structural space.
- Permutation-based null estimation.
- Benjamini–Hochberg FDR control.
- Significant region identification (FDR < 0.05).
- Regional testing and regional effect size (
regionalTestFC()).
Supported analysis modes demonstrated in this repository:
- One-dimensional (chain length or unsaturation)
- Two-dimensional (chain length × double bonds)
- Abundance-weighted and unweighted smoothing
- Odd–even chain stratification
data/ processed input data, one .rds per analysis (26 files)
scripts/ analysis scripts, mirroring the data layout (27 files)
results/ figures and tables produced by run_all.R
logs/ per-script output and run_summary.tsv
renv/ renv infrastructure; renv/cellar/ holds the LipidTrend tarball
docs/ repository diagram and input-data checksums
run_all.R run every analysis script
The diagram below summarizes the relationship between datasets, analysis modules, and manuscript figures.
All datasets analyzed in this study were obtained from previously published work and are described in the manuscript and Supplementary Table 3.
No raw data redistribution beyond published material is included.
The LipidTrend R package is publicly available:
- Bioconductor: https://bioconductor.org/packages/LipidTrend
- GitHub: https://github.com/BioinfOMICS/LipidTrend
This repository specifically documents the exact scripts and processed inputs used to generate the figures in the manuscript.
The analyses use LipidTrend 1.3.1, released through both channels above. The
same version is also bundled as a source tarball in renv/cellar/, so that
renv::restore() always installs precisely the version these results were
produced with — Bioconductor advances its release and devel branches on a
six-month cycle, and pinning the tarball keeps the environment reproducible
well beyond the lifetime of any one branch.
Releases of this repository are numbered after the LipidTrend version whose results they contain, so that the two can be matched at a glance.