Search

The nonhybrid hierarchical assembly of PacBio long reads is becoming the most preferred method for obtaining genomes for microbial isolates. On the other hand, among massive numbers of Illumina sequencing reads produced, there is a slim chance of re-evaluating failed microbial genome assembly (high contig number, large total contig size, and/or the presence of low-depth contigs). We generated Illumina-type test datasets with various levels of sequencing error, pretreatment (trimming and error correction), repetitive sequences, contamination, and ploidy from both simulated and real sequencing data and applied k-mer abundance analysis to quickly detect possible diagnostic signatures of poor assemblies. Contamination was the only factor leading to poor assemblies for the test dataset derived from haploid microbial genomes, resulting in an extraordinary peak within low-frequency k-mer range. When thirteen Illumina sequencing reads of microbes belonging to genera Bacillus or Paenibacillus from a single multiplexed run were subjected to a k-mer abundance analysis, all three samples leading to poor assemblies showed peculiar patterns of contamination. Read depth distribution along the contig length indicated that all problematic assemblies suffered from too many contigs with low average read coverage, where 1% to 15% of total reads were mapped to low-coverage contigs. We found that subsampling or filtering out reads having rare k-mers could efficiently remove low-level contaminants and greatly improve the de novo assemblies. An analysis of 16S rRNA genes recruited from reads or contigs and the application of read classification tools originally designed for metagenome analyses can help identify the source of a contamination. The unexpected presence of proteobacterial reads across multiple samples, which had no relevance to our lab environment, implies that such prevalent contamination might have occurred after the DNA preparation step, probably at the place where sequencing service was provided.