CYTO-SV-ML: A Machine Learning Tool for Cytogenetic Structural Variant Analysis in Somatic Cell Type Using Genome...

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Whole genome sequencing (WGS) enables comprehensive structural variant (SV) analysis, but distinguishing large somatic SVs (≥ 1 Mb) from germline SVs remains challenging.
• Accurate differentiation is crucial for diagnosing conditions like myelodysplastic syndromes (MDSs), where traditional cytogenetic methods have limitations.

Purpose of the Study:

• To develop and validate a machine learning pipeline (CYTO-SV-ML) for accurate identification of somatic cytogenetic SVs from WGS data.
• To characterize structural variation profiles in patients with MDSs using the developed pipeline.

Main Methods:

• A customized machine learning pipeline (CYTO-SV-ML) was developed using Snakemake, incorporating an automated workflow and user interface.
• An AUTO-ML model was trained and validated using known SVs from open databases.
• The pipeline was applied to whole blood WGS data from MDS patients.

Main Results:

• CYTO-SV-ML demonstrated high performance in classifying somatic cytogenetic SVs, with AUCROC values of 0.94 (translocations) and 0.92 (non-translocations).
• The pipeline identified 207 somatic cytogenetic SVs, outperforming a conventional SV calling pipeline (143 SVs) in validation against clinical records.
• Novel somatic cytogenetic SVs were discovered in 89% of MDS patients who had unsuccessful clinical cytogenetic results.

Conclusions:

• The CYTO-SV-ML pipeline offers a high-performance machine learning approach for classifying SVs from genomic sequencing data.
• Further validation of novel anomalies using orthogonal methods is necessary to realize the full clinical potential for cytogenetic diagnostics.