Pattern Recognition on Read Positioning in Next Generation Sequencing

The usefulness and the utility of the next generation sequencing (NGS) technology are based on the assumption that the DNA or cDNA cleavage required to generate short sequence reads is random. Several previous reports suggest the existence of sequencing bias of NGS reads. To address this question in greater detail, we analyze NGS data from four organisms with different GC content, Plasmodium falciparum (19.39%), Arabidopsis thaliana (36.03%), Homo sapiens (40.91%) and Streptomyces coelicolor (72.00%). Using machine learning techniques, we recognize the pattern that the NGS read start is positioned in the local region where the nucleotide distribution is dissimilar from the global nucleotide distribution. We also demonstrate that the mono-nucleotide distribution underestimates sequencing bias, and the recognized pattern is explained largely by the distribution of multi-nucleotides (di-, tri-, and tetra- nucleotides) rather than mono-nucleotides. This implies that the correction of sequencing bias needs to be performed on the basis of the multi-nucleotide distribution. Providing companion software to quantify the effect of the recognized pattern on read positioning, we exemplify that the bias correction based on the mono-nucleotide distribution may not be sufficient to clean sequencing bias.