bmVAE

bmVAE applies a variational autoencoder to learn low-dimensional representations of single-cell binary mutation data from single-cell DNA sequencing point mutations, enabling clustering of tumor cell subpopulations and genotype estimation to study genetic intra-tumor heterogeneity (ITH).

Key Features:

• Variational Autoencoder Framework: Uses a variational autoencoder (VAE) to learn latent representations from high-dimensional binary mutation matrices.
• Input Data: Operates on single-cell binary mutation data derived from single-cell DNA sequencing of point mutations.
• Dimensionality Reduction: Performs dimensionality reduction to compress high-dimensional binary mutation data while preserving structure relevant for downstream analysis.
• Cell Clustering: Clusters cells in the learned latent space to identify subpopulations corresponding to tumor clones.
• Genotype Estimation: Infers genotypes associated with each identified cell subpopulation.
• Handling Data Challenges: Explicitly addresses common single-cell mutation data issues including high false negative rates, variable dataset sizes, and intra-dataset heterogeneity.

Scientific Applications:

• Deciphering Genetic Intra-Tumor Heterogeneity (ITH): Facilitates analysis of tumor evolution and clonal diversity by clustering cells based on their mutational profiles.
• Personalized Cancer Therapy: Supports identification of distinct genetic subpopulations within tumors to inform tailored therapeutic strategies.

Methodology:

Accepts single-cell binary mutation matrices as input, processes them through a variational autoencoder for latent representation, performs clustering in the latent space and genotype estimation for each cluster, and has been validated on synthetic datasets simulating varying false negative rates, dataset sizes, and heterogeneity and benchmarked on real datasets.