sigLASSO

sigLASSO optimizes decomposition of somatic mutation counts into known mutational signatures using trinucleotide context while jointly modeling multinomial sampling likelihood to improve signature inference, particularly in low mutation count settings such as exome sequencing.

Key Features:

• Signature decomposition: Decomposes mutation counts into known mutational signatures based on trinucleotide context.
• Joint optimization: Integrates sampling likelihood with signature fitting into a unified objective function to account for multinomial sampling variance.
• L1 regularization (LASSO): Uses L1 regularization to produce sparse, interpretable signature assignments.
• Data-informed model complexity: Adjusts model complexity using data scale and biological priors and incorporates priors through soft thresholding.
• Model uncertainty assessment: Assesses model uncertainty to avoid assignments in low-confidence contexts.

Scientific Applications:

• Mutational signature deconvolution: Quantifies contributions of known mutational signatures in tumor genomes and exomes.
• Low-mutation-count analysis: Enables robust signature inference from exome sequencing and other datasets with few mutations by modeling multinomial sampling variance.
• Investigation of mutational processes: Supports inference of mutational processes and mechanisms of cancer development.

Methodology:

Jointly optimizes sampling likelihood and signature fitting in a unified objective function; explicitly models multinomial sampling variance and decomposes counts by trinucleotide context; applies L1 regularization for sparse solutions; adjusts model complexity using data scale and biological priors with soft thresholding; assesses model uncertainty to limit low-confidence assignments.