ImputeDistances

ImputeDistances imputes missing entries in phylogenetic distance matrices using machine learning to enable accurate phylogenetic and species tree estimation from incomplete genome-wide sequence data.

Key Features:

• Matrix Factorization Approach: Decomposes the distance matrix into factors that are used to predict missing distance entries.
• Autoencoder-Based Deep Learning Technique: Employs an autoencoder to model complex patterns in the distance matrix and reconstruct missing values.
• Handling Substantial Missing Data: Designed to impute distances when a significant fraction of the matrix entries are missing, outperforming many existing methods under such conditions.
• Scalability and Accuracy: The autoencoder-based approach emphasizes scalability for large datasets while both approaches report high accuracy in imputing distances.

Scientific Applications:

• Phylogenetic tree estimation: Provides imputed distance matrices for constructing more complete phylogenetic trees from incomplete data.
• Species tree estimation from genome-wide data: Facilitates species tree estimation by enabling use of genome-wide datasets that contain missing distance entries.

Methodology:

Performance is evaluated on simulated and biological datasets; the matrix factorization method decomposes the distance matrix into components to predict missing values; the autoencoder-based method uses deep learning to model and reconstruct incomplete distance matrices.