Clustering and graph mining techniques for classification of complex structural variations in cancer genomes

Area of Science:

• Genomics
• Cancer Biology
• Bioinformatics

Background:

• Distinguishing functional genomic variations from non-functional ones in cancer is a long-standing challenge.
• Complex chromosomal rearrangements (structural variants or SVs) are particularly difficult to classify and interpret functionally.

Purpose of the Study:

• To develop a robust statistical framework for classifying structural variants (SVs).
• To identify recurrent SV patterns indicative of specific molecular mechanisms in cancer.
• To analyze SVs across a large cohort of tumor samples.

Main Methods:

• Utilized a novel statistical approach incorporating recursive KDE clustering, randomization methods, and graph mining.
• Analyzed 152,926 SVs from 2392 tumor samples within the Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium.
• Applied statistical measures to identify significant recurrence and non-random occurrences of SV patterns.

Main Results:

• Successfully identified complex SV patterns across diverse cancer types.
• Demonstrated that these identified patterns are statistically significant and not random occurrences.
• Discovered a previously undescribed class of SV patterns.

Conclusions:

• The new statistical methodology provides a robust framework for classifying and interpreting structural variants in cancer genomics.
• The findings offer insights into the molecular mechanisms driving tumor biology by identifying recurrent, non-random SV patterns.
• The discovery of novel SV patterns opens new avenues for cancer research.