Bayclone: Bayesian nonparametric inference of tumor subclones using NGS data.

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Tumor heterogeneity (TH) is a key challenge in cancer research, impacting treatment response and disease progression.
• Accurate characterization of subclones and their evolutionary relationships is crucial for understanding cancer development.
• Existing models often struggle to capture the complex clonal architecture and overlapping mutations within tumors.

Purpose of the Study:

• To develop a novel feature allocation model for describing tumor heterogeneity using next-generation sequencing (NGS) data.
• To extend the Indian buffet process (IBP) using a Bayesian approach to create a categorical IBP (cIBP) for subclone analysis.
• To provide a more biologically realistic model that allows for overlapping mutations shared across subclones.

Main Methods:

• Developed a Bayesian nonparametric model, the categorical Indian buffet process (cIBP), extending the Indian buffet process (IBP).
• Defined subclones as vectors of categorical values representing genotypes at single nucleotide variations (SNVs).
• Employed a feature allocation approach, allowing somatic mutations to be shared across subclones, reflecting phylogenetic clonal expansion.

Main Results:

• The cIBP model successfully infers the number, genotypes, and proportions of subclones from NGS data.
• Bayesian inference provides posterior probabilities, enabling estimation of subclone characteristics and their variability.
• The model was validated on both simulated and real tumor sequencing data, demonstrating its efficacy.

Conclusions:

• The proposed cIBP model offers a powerful and flexible framework for dissecting tumor heterogeneity.
• This feature allocation approach provides a more accurate representation of clonal evolution compared to traditional clustering methods.
• The BayClone software implementation facilitates the application of this novel method in cancer genomics research.