WOLBACHIA STRAIN DIFFERENTIATION USING VARIATIONAL AUTOENCODERS

================================================================================WOLBACHIA STRAIN DIFFERENTIATION USING VARIATIONAL AUTOENCODERS================================================================================ This repository contains genomic alignments and VAE analysis results for Wolbachia strain differentiation described in: Corretto, E., Ragionieri, L., Wolfe, T.M., Palmieri, L., Serbina, L.S., Bruzzese, D.J., Klasson, L., Feder, J.L., Stauffer, C., and Schuler, H. (2026). Interspecific Horizontal Wolbachia Transmission Between Native and Invasive Fruit Flies: Tracing a Rapid Wolbachia Spread in Slow Motion. Current Biology. ================================================================================CONTENTS================================================================================ 1. wolbachia_alignment_47genes.phy - Original PHYLIP alignment of 47 genes across 44 Wolbachia samples - Format: PHYLIP (sequential) - Dimensions: 44 samples × 29,991 nucleotide positions 2. wolbachia_47genes_infection_newIDs.txt - One-hot encoded genomic sequences used for VAE training - Format: Tab-separated text - Column 1: Sample identifier - Column 2: Biological group (CER, CIN, SINGLE) - Columns 3+: One-hot encoded nucleotide vectors [A,C,G,T] 3. supervised_latent_coordinates.txt - VAE latent space coordinates from best training run - Format: Tab-separated text - Columns: sample, biological_group, latent_dim1, latent_dim2, silhouette_score, convergence_cv 4. supervised_results_summary.json - Complete numerical results and metrics from VAE analysis - Contains: training parameters, separation metrics, all run results 5. publication_figure_biological_groups_edited.pdf - Publication-quality figure showing strain separation in latent space ================================================================================BIOLOGICAL GROUPINGS================================================================================ Samples are classified into three groups based on infection status: - CER (n=23): Wolbachia from dual infections in R. cerasi hosts- CIN (n=21): Wolbachia from dual infections in R. cingulata hosts - SINGLE (n=6): Wolbachia from unusual single infections ================================================================================ANALYSIS SUMMARY================================================================================ Method: Two-stage Variational Autoencoder (VAE)- Stage 1: ~2000 epochs, KL weight = 0.3- Stage 2: ~2000 epochs, KL weight = 1.0- Latent dimensions: 2- Independent runs: 5- Best run silhouette score: 0.3620- Convergence CV: 3.77% The VAE successfully captured biologically meaningful genetic structure,with infection-based groupings showing strong separation in latent space. ================================================================================