SVModeler: Simulation of human haplotypes containing synthetic structural variants

Abstract

Abstract Motivation: Long-read sequencing overcomes the limits of short-reads by providing longer sequences that can span repetitive regions, improving the detection of structural variants and accurately resolving their sequences. Nonetheless, detection and annotation of structural variants remains a computational challenge, requiring active development and benchmarking of available algorithms. The current availability of detailed sequence information for large collections of SVs identified using long-read sequencing presents a valuable opportunity for developing and training realistic novel simulation frameworks, which can be used for the evaluation of SV callers. Results: SVModeler is a newly developed computational tool to simulate synthetic human haplotypes containing embedded SVs. A unique feature of SVModeler is its capability to leverage SV catalogs to model the genome-wide distribution, frequency and sequence features of various SV classes, including tandem duplications, mobile elements and variable number tandem repeats. As a proof of principle, SVModeler has been trained with a large catalog of polymorphic SVs identified in a dataset comprising 1.019 samples from the 1000 Genomes Project, which represents the largest collection of diverse humans sequenced with long reads to date. Code availability: https://github.com/ismaelveramu/SVModeler Contact: ismael.vera@uvic.cat