An exploration of assembly strategies and quality metrics on the accuracy of the rewarewa (Knightia excelsa) genome.

We used long read sequencing data generated from Knightia excelsa, a nectar producing Proteaceae tree endemic to Aotearoa (New Zealand), to explore how sequencing data type, volume and workflows can impact final assembly accuracy and chromosome reconstruction. Establishing a high-quality genome for this species has specific cultural importance to Māori and commercial importance to honey producers in Aotearoa. Assemblies were produced by five long read assemblers using data subsampled based on read lengths, two polishing strategies, and two Hi-C mapping methods. Our results from subsampling the data by read length showed that each assembler tested performed differently depending on the coverage and the read length of the data. Subsampling highlighted that input data with longer read lengths but perhaps lower coverage constructed more contiguous, kmer and gene complete than short read length input data with higher coverage. The final genome assembly was constructed into 14 pseudo-chromosomes using an initial FLYE long read assembly, a Racon/Medaka/Pilon combined polishing strategy, SALSA2 and AllHiC scaffolding, Juicebox curation, and Macadamia linkage map validation. We highlighted the importance of developing assembly workflows based on the volume and read length of sequencing data and established a robust set quality metrics for generating high quality assemblies. Scaffolding analyses highlighted that problems found in the initial assemblies could not be resolved accurately by Hi-C data and that assembly scaffolding was more successful when the underlying contig assembly was of higher accuracy. These findings provide an insight into how quality assessment tools can be implemented throughout genome assembly pipelines to inform the de novo reconstruction of a high quality genome assembly for non-model organisms.