Life at the extremes

This repository contains the official implementation of the study: Life at the extremes: Maximally divergent microbes with similar genomic signatures linked to extreme environments (Safari et al., 2025): preprint

In that work, we showed that extremophiles — despite belonging to maximally divergent lineages — can converge toward highly similar genomic k-mer signatures when adapting to extreme environments (temperature, pH, and beyond). These convergent patterns highlight the role of large-scale mutational and selective pressures in shaping microbial genomes under stress.

---

Installation

# From repo root
pip install -e .

Download Data

Download the extremophile genome assemblies and metadata from Zenodo DOI: link

Data Layout

Place downloaded assemblies (FASTA .fna) and the metadat file under data/.

All results will be written to outputs/ automatically.

All the results of the experiments of this study are available in results/ folder.

Supervised Learning Experiments

1) Effect of Genome Proxy Selection (Multiple Runs)

Tests whether random genome proxy choice changes classification accuracy.

python3 src/extprime/pipelines/pipeline_supervised.py \
  --exp_type exp1 --max_k 6 --data_root data --output_root outputs

2) Accuracy vs. Genome Proxy Length (Single Run)

Compares accuracy across proxy lengths (k set by --max_k).

python3 src/extprime/pipelines/pipeline_supervised.py \
  --exp_type exp2 --max_k 6 --data_root data --output_root outputs

3) Effect of Number of Subfragments (n)

Varies n in the composite genome proxy to measure its impact.

python3 src/extprime/pipelines/pipeline_supervised.py \
  --exp_type exp3 --max_k 6 --data_root data --output_root outputs

Optional: Add --whole_genome to use entire genomes instead of proxies.

Outputs

Results are written under:

outputs/{exp_type}/{env}/fragments_{length}/...

Each folder contains the generated FASTA for that environment and the model outputs produced by the pipeline.

Command-Line Flags

• --exp_type {exp1,exp2,exp3,tuning} – Choose the experiment
• --max_k INT – Maximum k-mer length considered by the models
• --data_root PATH – Input root (default: data)
• --output_root PATH – Results root (default: outputs)
• --whole_genome – Use entire genomes instead of proxies (optional)

Unupervised Learning Experiments

4) Non-parametric clustering and candidate identification

python3 src/extprime/pipelines/pipeline_unsupervised.py \
  --exp_type non-parametric --k_mer 6 --data_root path_to_the_subfragments --output_root outputs \
  --fragement_length 100000 --n_clusters 4 --env Temperature

Command-Line Flags

• --exp_type {parametric, non-parametric} – Choose the experiment
• --k_mer INT – K-mer length
• --fragement_length INT – Fragment length
• --n_clusters INT – Number of clusters (default: 4) - not needed for non-parametric
• --outputs_root PATH – Output results directory
• --env {pH,Temperature} – Environment type
• --fragment_path PATH – Path to a fragment FASTA file
• --data_root PATH – Input data directory

FCGR distance calculation and filtering

python3 src/extprime/analysis/distance_calculator.py --data_root path_to_the_subfragments