SUITOR

SUITOR selects the optimal number of mutational signatures through unsupervised cross-validation by splitting a mutational counts data set into training and validation data, estimating signatures in the training data, and using them to predict mutations in the validation data.

Key Features:

• Unsupervised cross-validation: SUITOR requires little assumptions and no numerical approximations to select the optimal number of signatures without overfitting the data.
• Training-validation split: A data set of mutational counts is split into training and validation data for signature estimation and prediction.
• Signature prediction: Signatures estimated in the training data are used to predict the mutations in the validation data.

Scientific Applications:

• Cancer genomics studies: SUITOR was applied to 2,540 whole-genome sequenced tumors across 22 cancer types and selected signatures with the smallest prediction errors.
• Breast cancer signature validation: Almost all signatures of breast cancer selected by SUITOR were validated in an independent breast cancer study.
• De novo mutational signature analysis: SUITOR facilitates downstream analyses with etiological or therapeutic importance by correctly identifying signatures, some of which were missed by other widely used methods.

Methodology:

SUITOR splits a data set of mutational counts into training and validation data, estimates signatures in the training data, and uses them to predict the mutations in the validation data.