Preprocessing make critical bias at microbial community

Abstract

Next-generation sequencing technology (NGS) is becoming a standard method to examine microbial diversity of environmental and human microbiome samples. Read number is used to estimate relative abundance of specific taxa in the samples as a key parameter to define microbial community structures. However, read number is affected not only by relative abundance of the taxa in the original sample but also by various processes such as PCR amplification, sequencing reactions and sequence processing pipelines including trimming, filtering, noise removing, chimera detection, clustering, and DB search. Although sequence read number biases by PCR and sequencing reactions are well known, read number bias created by quality trimming and filtering is hardly known. Trimming low-quality nucleotides and filtering out short sequence reads are included in most of the popular NGS processing pipelines to analyze microbial community based on only high-quality sequence reads. These processes are based on the assumption that low-quality reads are evenly distributed among taxa. However, we found that sequence quality is highly dependent on the sequence context, which in turn can create biased sequence trimming and discarding sequences from specific taxa. Therefore, we propose that sequence quality trimming and filtering should be conducted with more lenient trimming options or trimming read at the last high-quality region instead of trimming at the first low-quality region/base to mitigate read number bias.