Analysis of High-throughput sequencing (HTS) data is a difficult problem, especially in the context of non-model organisms where comparison of homologous sequences may be hindered by the lack of a close reference genome. Current mapping-based methods rely on the availability of a highly similar reference sequence, whereas de novo assemblies produce anonymous (unannotated) contigs that are not easily compared across samples. Here, we present Assembly by Reduced Complexity (ARC) a hybrid mapping and assembly approach for targeted assembly of homologous sequences. ARC is an open-source project (http://ibest.github.io/ARC/) implemented in the Python language and consists of the following stages: 1) align sequence reads to reference targets, 2) use alignment results to distribute reads into target specific bins, 3) perform assemblies for each bin (target) to produce contigs, and 4) replace previous reference targets with assembled contigs and iterate. We show that ARC is able to assemble high quality, unbiased mitochondrial genomes seeded from 11 progressively divergent references, and is able to assemble full mitochondrial genomes starting from short, poor quality ancient DNA reads. We also show ARC compares favorably to de novo assembly of a large exome capture dataset for CPU and memory requirements; assembling 7,627 individual targets across 55 samples, completing over 1.3 million assemblies in less than 78 hours, while using under 32 Gb of system memory. ARC breaks the assembly problem down into many smaller problems, solving the anonymous contig and poor scaling inherent in some de novo assembly methods and reference bias inherent in traditional read mapping.

Assembly by Reduced Complexity (ARC): a hybrid approach for targeted assembly of homologous sequences.