Variational Autoencoder Implementation for Geospatial-Genetic Analysis of Coptis chinensis

This repository contains a Python implementation of VAE-based geospatial-genetic analysis for Coptis chinensis populations. Pipeline:VCF genotype data → 4n-dimensional features → VAE (10D latent space) → PCA (PC1) → Kriging interpolation → Geospatial maps Key Components: Variational Autoencoder with architecture: 4n → 512 → 256 → 10 → 256 → 512 → 4n4n-dimensional feature construction: genotypes, allele frequencies, and genomic positionsOrdinary Kriging spatial interpolation with exponential variogramReproducible analysis with fixed random seed (seed=42)Input Requirements: VCF file with genotype dataExcel file with sample coordinates (longitude/latitude)Outputs: Trained VAE model weightsLatent representations and PC1 valuesHigh-resolution geospatial distribution maps (300 DPI)Complete analysis report with metadataSoftware: Python ≥3.7, PyTorch, scikit-allel, pykrige, pandas, numpy, matplotlib