Characterization of sequence-specific errors in various next-generation sequencing systems

Next-generation sequencing (NGS) is a popular method for assessing the molecular diversity of microbial communities without cultivation, for identifying polymorphisms in populations, and for comparing genomes and transcriptomes. However, sequence-specific errors (SSEs) by NGS systems can result in genome mis-assembly, overestimation of diversity in microbial community analyses, and false polymorphism discovery. SSEs can be particularly problematic due to rich microbial biodiversity and genomes containing frequent repeats. In this study, SSEs in public data from all popular NGS systems were discovered using a Markov chain model and hotspots for sequence errors were identified. Deletion errors were frequently preceded by homopolymers in non-Illumina NGS systems, such as GS FLX+. Substitution errors were often related to high GC contents and long G/C homopolymers in Illumina sequencing systems such as HiSeq. After removal of long G/C homopolymers in HiSeq, the average lengths of contigs and average SNP quality increased. SSEs were selectively removed from our mock community data by quality filtering, and a bias against specific microbes was identified. Our findings provide a scientific basis for filtering poor-quality reads, correcting deletion errors, preventing genome mis-assembly, and accurately assessing microbial community compositions and polymorphisms.