Phenotypic geometry is heterogeneous across the biological hierarchy

This repository contains code accompanying 'Phenotypic geometry is heterogeneous across the biological hierarchy'. These analyses characterize the dimensionality, continuity, topology, curvature, ancestry dependence, heritability, and cross-layer recoverability of RNA abundance, protein abundance, and growth phenotypes in 802 natural Saccharomyces cerevisiae isolates.

Installation

Install Miniconda and Mamba, then create the pinned R 4.3 environment:

mamba env create --name phenogeometry --file envs/dev.yml
conda activate phenogeometry
Rscript -e 'remotes::install_version("TDAstats", version = "0.4.1", repos = "https://cloud.r-project.org")'

The workflow uses base R plus ape, diptest, igraph, lpSolve, mclust, readxl, and TDAstats. TDAstats is installed from CRAN because it is not distributed as a Conda package. No GPU is required.

Data

The repository includes the following inputs:

• data/genotypes/1011DistanceMatrixBasedOnSNPs.tab: SNP-distance matrix for the 1,011-isolate panel from Peter et al. (2018).
• data/genotypes/strain_clades_peter2018_S1.txt: major-clade assignments from Peter et al. (2018).
• data/41586_2018_30_MOESM3_ESM.xls: Peter et al. (2018) strain metadata used for the phylogeny display.
• data/phenotypes/loegler_phenos_cleaned_zscore.RDS: cached cleaned analysis matrices so downstream scripts can be run without repeating primary cleaning.

code/00_load_and_clean_data.R downloads Phenotypes_8391Traits.tar.gz from Loegler et al.'s Zenodo record when the source directory is absent and verifies its published MD5 checksum before extraction. Downloaded source files are Git-ignored.

The cleaning script reconstructs the cached object and its companion files from those source files. It intersects strains across layers, removes strains with more than 20% missing growth measurements, removes traits with more than 20% missingness, mean-imputes remaining values, and z-scores each trait. The resulting matrices contain 802 strains and 5,970 RNA, 630 protein, and 241 growth traits.

The generated companion files are:

• loegler_phenos_cleaned.RDS: filtered and mean-imputed matrices on the original measurement scales.
• loegler_phenos_cleaned_zscore.RDS: standardized matrices used by the analyses.
• loegler_phenos_raw.RDS: NA-preserving matrices on the retained strain set, before trait filtering and imputation.
• loegler_strains_cleaned.RDS: retained strain identifiers.

Repository structure

code/       ordered R scripts for downloading, cleaning, analyses, and figures
data/       compact inputs, generated phenotype matrices, and ignored downloads
envs/       Conda environment specification
results/    serialized analysis results
figures/    generated manuscript-facing source panels (created on run)

Reproducing the analyses

Run commands from the repository root. To regenerate every cleaned data object, result, and script-generated figure panel in order:

Rscript code/99_run_paper_analyses.R

Each script can also be run independently in the order below. Scripts use seed 42 unless noted, write serialized results to results/, and write graphics to figures/.

Order	Script	Manuscript output
00	00_load_and_clean_data.R	Primary phenotype cleaning and cached matrices
01	01_setup.R	Shared paths, loaders, estimators, and plotting helpers; sourced rather than run directly
02	02_tree_measurements.R	Figure 1 phylogeny and measurement rings
03	03_dimensionality.R	Figures 2–3 dimensionality and strain-rarefaction results
04	04_continuum.R	Figure 4 dip tests, GMM/BIC sweep, and Hopkins comparisons
05	05_curvature.R	Figure 5 curvature calibration and phenotype diagnostics
06	06_orthogonal_geometry.R	Persistent homology plus exploratory orthogonal geometry checks; results only
07	07_h0_persistence_nulls.R	Manuscript-facing H0 persistence null analysis
08	08_population_structure_figure.R	Figure 6 ancestry, heritability, and tree-cut analyses
09	09_variance_partition.R	Figure 7A cross-layer variance partition
10	10_variance_partition_figure.R	Figure 7A rendered source panel
11	11_cross_layer_geometry.R	Figure 7B–C joint PR and pairwise geometry panels; prediction and sufficiency results

The scripts write only panels used in the paper. Accordingly, the repository reproduces the underlying analyses and script-generated source panels but does not claim pixel-identical regeneration of those three final composites.

Running selected stages

For example, the Figure 7 workflow is:

Rscript code/09_variance_partition.R
Rscript code/10_variance_partition_figure.R
Rscript code/11_cross_layer_geometry.R

Set PG_ROOT to use a data directory outside the repository, PG_OUT to redirect results/ and figures/, or PG_PHENO_OUT to redirect the cleaned phenotype objects. PG_PHENO_URL can point to an exact mirror of the checksum-verified source archive.

PG_ROOT=/path/to/input-root PG_OUT=/path/to/output-root \
  Rscript code/11_cross_layer_geometry.R

Results

Each analysis saves an RDS file with the same numeric prefix and descriptive stem as its generating script. Existing results in results/ are release snapshots.

Compute specifications

The release workflow is CPU-only and runs as serial R scripts. It was developed on a MacBook Pro running macOS 15.7.4 and R 4.3.2.

License and contributing

Code is released under the MIT License. For feedback and contributions, see Arcadia Science's guide to recognizing software contributions.