Using sigLASSO to optimize cancer mutation signatures jointly with sampling likelihood.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Carcinogenesis involves multiple mutational processes that leave distinct signatures in tumor genomes.
• Identifying active mutational signatures is crucial for understanding cancer development mechanisms.

Purpose of the Study:

• To develop an efficient software tool, sigLASSO, for decomposing cancer mutation counts into known signatures.
• To improve the accuracy and interpretability of mutational signature analysis.

Main Methods:

• Developed sigLASSO, a software tool implementing L1 regularization for parsimonious signature assignment.
• Jointly optimized sampling likelihood and signature fitting by factoring multinomial sampling.
• Incorporated data scale and biological priors to fine-tune model complexity.

Main Results:

• sigLASSO provides sparse and interpretable solutions for mutational signature decomposition.
• The tool efficiently handles low mutation counts and high sampling variance, common in exome sequencing.
• sigLASSO assesses model uncertainty and abstains from low-confidence assignments.

Conclusions:

• sigLASSO offers an efficient and robust method for identifying cancer mutational signatures.
• The tool enhances the understanding of cancer development by providing interpretable genomic insights.
• sigLASSO's ability to assess uncertainty improves the reliability of mutational signature analysis.