TRONCO

TRONCO reconstructs causal models of cancer progression from (epi)genomic mutational profiles, implemented as an R package to infer population- and individual-level evolutionary trajectories.

Key Features:

• Population-Level Models: Extracts trends in the accumulation of genetic alterations across cohorts and supports analysis of datasets such as The Cancer Genome Atlas (TCGA).
• Individual-Level Models: Reconstructs clonal evolutionary histories for single patients with multiple samples, including multi-region biopsies and single-cell sequencing data.
• Probabilistic Causation Framework (CAPRESE): Implements CAPRESE based on Suppes' notion of probabilistic causation and employs shrinkage-like estimators to enhance noise tolerance.
• Directed Acyclic Graphs (DAGs): Represents cancer progression models as DAGs encoding 'selectivity' relations between mutations.
• CAPRI Algorithm and Performance: Implements the CAPRI algorithm for cancer progression inference, demonstrating robust handling of noisy data and limited sample sizes with efficient convergence as sample numbers increase.
• Validation and Comparative Inference: Facilitates validation of conjectured selective relations and comparison of progression models inferred from different datasets and sequencing technologies.

Scientific Applications:

• Cohort-Level Evolutionary Analysis: Infers cohort-level progression trajectories from cross-sectional genomic data such as TCGA to study cancer evolution at the population scale.
• Patient-Specific Clonal Reconstruction: Reconstructs patient-specific clonal evolution from multi-region biopsies or single-cell sequencing for personalized analyses.
• Biological Hypothesis Validation: Enables testing and validation of conjectured biological relationships among mutations within inferred progression models.
• Stratification and Comparative Studies: Supports stratification of patients and comparison of progression patterns across datasets generated by different sequencing technologies.

Methodology:

Implements CAPRESE and CAPRI algorithms using a scoring framework based on Suppes' probabilistic theory, incorporates shrinkage-like estimators for noise robustness, and applies bootstrap and maximum likelihood inference techniques.