SUITOR

SUITOR estimates the optimal number of de novo mutational signatures in cancer genomics datasets using an unsupervised cross-validation framework to guide signature selection.

Key Features:

• De novo mutational signature analysis: Performs de novo extraction of mutational signatures from mutation catalogs.
• Optimal signature number selection: Determines the optimal number of mutational signatures present in a dataset.
• Unsupervised cross-validation: Employs an unsupervised cross-validation methodology to evaluate signature models.
• Minimizes assumptions: Minimizes methodological assumptions during cross-validation.
• Avoids numerical approximations: Avoids numerical approximations to reduce potential estimation bias.
• Prediction-error minimization: Selects signatures by identifying models with the smallest prediction errors.
• Overfitting reduction: Reduces the risk of overfitting through its cross-validation-based selection criterion.
• Validation in vitro and in silico: Performance has been validated using in vitro experiments and in silico simulations.
• Application to whole-genome sequencing: Has been applied to 2,540 whole-genome sequenced tumors across 22 cancer types.
• Independent breast cancer corroboration: Identified signatures were corroborated in an independent breast cancer study.

Scientific Applications:

• Etiological inference: Supports inference of mutational processes underlying cancer development.
• Therapeutic implication analysis: Facilitates investigation of potential therapeutic implications of mutational signatures.
• Large-cohort signature discovery: Enables signature discovery and selection in large whole-genome sequencing cohorts, including multi-cancer and breast cancer studies.
• Downstream analytical accuracy: Improves accuracy of downstream analyses that depend on correct signature number estimation.

Methodology:

Performs de novo mutational signature extraction combined with an unsupervised cross-validation framework that minimizes assumptions, avoids numerical approximations, and selects the number of signatures by minimizing prediction error.