Phylogenetic trees inferred with different algorithms dataset and their SPR comparison

Dataset Organization This dataset contains 3 different datasets: 1) dataset with input data for phylogenetic tree inference (Profile Dataset) 2) dataset with trees inferred from different phylogenetic methods (Phylogenetic Tree Dataset) 3) dataset with the topological differences between phylogenetic trees, quantified through the approximated SPR (Subtree Prune and Regraft) metric (SPR Dataset) using the phangorn R package. 1. Profile Dataset This repository contains organized datasets for the processing and analysis of genomic sequences from 4 distinct bacterial species. The data is structured into different processing phases (raw, prepared, and shuffled) and organized by dimension. 1.1. Directory Structure The dataset is divided into three main directories: full_datasets/: Contains the original files (raw) for each species for cgmlst (obtained in April, 2026) until a given size k (first line to line k). clostridium.txt from Enterobase database (https://enterobase.warwick.ac.uk/species/clostridium/download_data) pneumonia.txt from PUBMLST database (https://pubmlst.org/bigsdb?db=pubmlst_spneumoniae_isolates&page=plugin&name=Export) salmonella.list from Enterobase database (https://enterobase.warwick.ac.uk/species/senterica/download_data) vibrio.txt from Enterobase database (https://enterobase.warwick.ac.uk/species/vibrio/download_data) prepared_datasets/: Processed data segmented by size. shuffled_datasets/: Data versions with shuffled sequences (with the _sh.txt suffix), maintaining the same categorization by size. 1.2. Dimension Categorization The datasets in the directories prepared_datasets and shuffled_datasets are divided into 3 categories, based on the number of sequences (from the first line to the k/dimension line): Category Dimension Range (Sequences) Small 90 – 1,000 Medium 2,000 – 6,000 Large 7,000 – 9,500 1.3. Included Species The dataset covers the following bacterial species: Vibrio Clostridium Salmonella Klebsiella pneumoniae (referenced as pneumonia) 1.4. File Nomenclature The files follow a naming convention that identifies the species and the exact dimension: Standard Format: {species}_{size}.txt Example: clostridium_9000.txt (Large category file) Shuffled Format: {species}_{size}_sh.txt Example: clostridium_7000_sh.txt (Shuffled version of a 7,000-sequence dataset) 1.5 Processing Notes Files are in plain text format (.txt or .list). The shuffled_datasets folder is intended for robustness testing or model training where sequence order should not influence results. 2. Phylogenetic Trees Dataset Organization This directory contains phylogenetic trees generated for various bacterial species, organized to facilitate the calculation of SPR (Subtree Prune and Regraft) metrics based on data size and the specific construction algorithm used. There are 864 trees. 2.1 Directory Structure The main directories are divided by the following microorganisms: clostridium(https://enterobase.warwick.ac.uk/species/index/clostridium) vibrio (https://enterobase.warwick.ac.uk/species/index/vibrio) salmonella (https://enterobase.warwick.ac.uk/species/index/senterica) pneumonia (https://pubmlst.org/organisms/streptococcus-pneumoniae) Inside each of these directories, files are categorized into three different size tiers: Category Size Range (Isolates/Sequences) Small 99 to 1,000 Medium 2,000 to 6,000 Large 6,000 to 9,5000 2.2. Methodology and Algorithms For each dataset, trees were generated using two of the most established algorithms in phylogenetics: 2.2.1. UPGMA (Unweighted Pair Group Method with Arithmetic Mean) UPGMA is a simple hierarchical clustering method. It assumes a constant rate of evolution (the molecular clock hypothesis) and constructs ultrametric trees, where all leaves are equidistant from the root. 2.2.2. NJ (Neighbor-Joining) Neighbor-Joining is a distance-matrix method that does not assume a constant rate of evolution. It is widely used for its computational efficiency and its ability to handle varying branch lengths, making it effective at reconstructing topologies where evolutionary rates differ across lineages. 2.3.Tools Used All trees in this repository were generated using the phyloLib library. phyloLib is a robust tool for processing and manipulating phylogenetic data. You can find the official repository here: https://github.com/phyloviz/phylolib 3. SPR Dataset Organization This directory contains the processed datasets designed to be used as inputs for Graph Neural Networks (GNN). These files represent the topological differences between phylogenetic trees, quantified through the aproximated SPR (Subtree Prune and Regraft) metric using the phangorn R package. 3.1. Directory Structure The files (e.g., spr_metrics_clostridium_small.csv) follow a specific naming convention based on the microorganism and the size tier (Small, Medium, or Large). CSV File Format The data is structured in columns to facilitate batch processing for GNN training: Model 1: The first tree in the comparison (e.g., UPGMA). Model 2: The second tree in the comparison (e.g., NJ). SPR Distance: The raw SPR distance between the two trees. 3.2. Methodology & SPR Algorithm To analyze the structural divergence between algorithms, we calculate the SPR (Subtree Prune and Regraft) distance. This metric counts the minimum number of operations required to transform one tree topology into another by pruning a subtree and reattaching it elsewhere. This allows the GNN to learn the "topological gap" between different reconstruction methods for the same biological data. 3.3. Tools & Libraries The trees and the corresponding SPR metrics were generated using the phangorn R package. phangorn is a specialized library for phylogenetic analysis, offering advanced methods for estimating phylogenies and comparing tree distances. Package Reference: CRAN - phangorn