Substantial biases in ultra-short read data sets from high-throughput DNA sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Next-generation sequencing

Background:

• Novel sequencing technologies enable rapid, large-scale data production, poised to transform genetics and biomedical research.
• Thorough characterization of ultra-short read output is crucial for reliable data interpretation.

Purpose of the Study:

• To analyze error rates, biases, and quality scores in ultra-short read data from Illumina 1G sequencing.
• To assess the impact of sequencing errors and biases on various genomic applications.

Main Methods:

• Generated and analyzed two ultra-short read datasets: 2.8 million 27mer reads (Beta vulgaris) and 12.3 million 36mer reads (Helicobacter acinonychis).
• Evaluated error rates, base substitution frequencies, insertion/deletion rates, and read coverage biases.
• Simulated re-sequencing to determine sufficient coverage for error compensation.
• Assessed the accuracy of Solexa quality scores.

Main Results:

• Error rates varied from 0.3% to 3.8%, with errors often preceded by base G.
• Specific base substitution patterns were identified, with A>C transversions being frequent and C>G transversions infrequent.
• Single base insertions/deletions occurred at low rates; 20-fold coverage was sufficient to overcome errors.
• Read coverage exhibited bias, correlating with GC content; Solexa quality scores were found to be unreliable.

Conclusions:

• Ultra-short read sequencing data contain systematic biases and errors that necessitate careful detection and interpretation.
• Understanding these characteristics is vital for accurate de novo sequencing, re-sequencing, SNP identification, and transcriptome analysis.