ConnectedReads: machine-learning optimized long-range genome analysis workflow for next-generation sequencing

Abstract

Current human genome sequencing assays in both clinical and research settings primarily utilize short-read sequencing and apply resequencing pipelines to detect genetic variants. However, theses mapping-based data analysis pipelines remains a considerable challenge due to an incomplete reference genome, mapping errors and high sequence divergence. To overcome this challenge, we propose an efficient and effective whole-read assembly workflow with unsupervised graph mining algorithms on an Apache Spark large-scale data processing platform called ConnectedReads. By fully utilizing short-read data information, ConnectedReads is able to generate assembled contigs and then benefit downstream pipelines to provide higher-resolution SV discovery than that provided by other methods, especially in high diversity against reference and N-gap regions of reference. Furthermore, we demonstrate a cost-effective approach by leveraging ConnectedReads to investigate all spectra of genetic changes in population-scale studies.