Machine learning-predicted chromatin organization landscape across pediatric tumors

Area of Science:

• Genomics
• Cancer Biology
• Computational Biology

Background:

• Structural variants (SVs) significantly impact oncogenesis by altering 3D genome folding.
• Whole-genome sequencing enables large-scale SV profiling, but experimental validation of their functional impact is challenging.

Purpose of the Study:

• To predict and analyze SV-induced genome folding disruptions across diverse pediatric tumor types using machine learning.
• To identify recurrently disrupted genomic regions and functional mechanisms associated with SVs in cancer.

Main Methods:

• Utilized a convolutional neural network (Akita) to predict genome folding disruptions from somatic SVs in 61 tumor types.
• Applied dimensionality reduction to SV disruption scores to identify key disrupted regions.
• Integrated epigenetic data and an Activity-by-Contact scoring approach to prioritize SVs affecting active enhancers.

Main Results:

• Demonstrated significant variability in SV-induced disruptions across tumor types, with lymphomas, sarcomas, and germline cell tumors showing the most disruptive SVs.
• Identified five recurrently disrupted regions enriched for high-impact SVs, some harboring tumor-associated genes and regulators.
• Highlighted disruptive SVs near oncogenes and identified novel candidate loci using epigenetic data integration.

Conclusions:

• Machine learning effectively predicts SV-driven genome folding alterations in pediatric cancers.
• This approach identifies novel cancer-associated loci and mechanisms, providing a framework for future research on SVs in tumorigenesis.