SSLCNV: A Semi-supervised Learning Framework for Accurate Copy Number Variation Detection

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Copy number variation (CNV) is a key structural variation impacting genetic diversity and disease.
• Existing CNV detection tools have limitations in sensitivity, breakpoint resolution, and robustness in complex sequencing data.
• There is a need for improved CNV detection methods that leverage existing tool strengths.

Purpose of the Study:

• To develop a novel semi-supervised learning framework (SSLCNV) for accurate CNV detection.
• To enhance CNV detection accuracy and robustness, particularly in complex sequencing environments.
• To improve the utilization of existing CNV detection tools.

Main Methods:

• SSLCNV employs consensus-based pseudo-labeling by intersecting predictions from four representative CNV tools.
• Density-based clustering with a novel constraint z-score is used to enhance accuracy.
• The framework effectively detects CNVs from partially labeled and unlabeled data.

Main Results:

• SSLCNV consistently achieves superior F1-scores compared to existing tools across various sequencing depths and tumor purities.
• The method demonstrates robust performance under low-coverage conditions, improving recall without significant precision loss.
• Comprehensive evaluations on simulated and real datasets validate SSLCNV's effectiveness.

Conclusions:

• SSLCNV provides a scalable and accurate solution for CNV detection.
• The framework is particularly advantageous for complex genomic backgrounds and low-coverage data.
• SSLCNV represents a significant advancement in leveraging multiple tools for improved CNV identification.