Chromothripsis detection with multiple myeloma patients based on deep graph learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Chromothripsis is a significant genomic event linked to poor outcomes in multiple myeloma.
• Early detection of chromothripsis aids in risk estimation and treatment strategies for multiple myeloma patients.
• Current detection methods rely on manual analysis of whole-genome sequencing data, including structural variations, which is resource-intensive.

Purpose of the Study:

• To develop a reliable and accurate method for detecting chromothripsis using only copy number variation (CNV) data.
• To reduce the dependency on expert manual diagnosis and structural variation data extraction.
• To facilitate earlier and more accessible risk assessment in multiple myeloma.

Main Methods:

• Proposed a novel method for chromothripsis detection based exclusively on CNV data.
• Employed structure learning to infer a directed acyclic graph (DAG) of CNV features, creating a CNV embedding graph (CNV-DAG).
• Utilized a Graph Transformer-based neural network incorporating local feature extraction and non-linear feature interaction for event classification.

Main Results:

• The proposed method successfully detects chromothripsis events using only CNV data.
• Ablation experiments, clustering, and feature importance analysis validated the model's performance and interpretability.
• The approach offers a more accessible alternative to traditional methods requiring structural variation data.

Conclusions:

• The developed CNV-based method provides an accurate and efficient approach for chromothripsis detection in multiple myeloma.
• This method has the potential to improve risk stratification and inform early treatment decisions.
• The open-source availability of the code and data promotes further research and clinical application.