BnpC

BnpC performs non-parametric Bayesian clustering of single-cell DNA sequencing (scDNA-seq) data to infer clonal populations and their genotypes from noisy mutation profiles for studies of intra-tumor heterogeneity.

Key Features:

• Non-Parametric Clustering: Uses a Dirichlet process mixture model to infer an unknown number of clonal populations from scDNA-seq mutation profiles.
• MCMC Sampling Scheme: Employs a Markov Chain Monte Carlo framework with Gibbs sampling, modified split-merge moves, and Metropolis-Hastings updates to explore the joint posterior of clustering and genotypes.
• Posterior Estimator: Implements a posterior estimator that accounts for posterior shape to improve clone and genotype prediction from noisy and incomplete data.

Scientific Applications:

• Intra-tumor Heterogeneity Analysis: Infers clonal composition and genotypes from scDNA-seq to resolve genetic heterogeneity within tumors.
• Clonal Genotype Identification: Distinguishes clonal populations based on mutation profiles to identify genetic variants relevant to progression or treatment response.
• High-throughput scDNA-seq Processing: Scales to large scDNA-seq datasets (reported up to ~10,000 cells) for population-level clonal inference and downstream analyses.

Methodology:

Inference is based on a Dirichlet process mixture model sampled via MCMC using Gibbs sampling supplemented by modified split-merge moves and Metropolis-Hastings updates, with a posterior estimator summarizing clones and genotypes.