Structural Variant Calling in Genomes Using Deep Learning
Completed:
This thesis explores the integration of deep learning techniques to improve the detection of structural variants (SVs) in the human genome, leveraging long-read sequencing data. Building on the PEPPER-Margin-DeepVariant framework, it introduces novel methodologies to detect and cluster SVs using normalized indel similarity scores and soft-clip analysis. The approach demonstrates superior performance on benchmark datasets, achieving state-of-the-art F1 scores of 98.87% on CMRG and 96.66% on HG002, thereby establishing a new paradigm for accurate and efficient SV detection.