SUGAR: graphical user interface-based data refiner for high-throughput DNA sequencing

Next-generation sequencers (NGSs) have become one of the main tools for current biology, being used in the analyses of genomic sequences, gene expressions, and promoter activities. To extract valuable insights from NGSs, it is essential to obtain high quality sequencing data. However, the quality is sometimes affected by errors during sequencing procedures. In the Illumina HiSeq and MiSeq machines, experimental small problems like air bubbles and water condensation in the flowcell are known to reduce sequencing qualities [1].

It is usually difficult to avoid accidental and stochastic occurrence of such technical errors in sequencing procedures. Even though on average the obtained sequences were high-quality, it may also contain low-quality sequences generated from physical portions of the flowcell (tiles) affected by the technical errors. Finding and cleaning such error-affected tiles from the whole data would be useful to obtain reliable results and insights from the NGS data analyses.

Here, we develop a graphical user interface (GUI)-based software SUGAR (subtile-based GUI-assisted refiner). This enables rapid evaluation and cleaning of the Illumina HiSeq and MiSeq data, specifically considering technical errors in flowcells and sequencing run. Novelty of this software includes three points. First, SUGAR is capable of analyzing whole data generated by ultra-high-throughput HiSeq machine. The full data of a HiSeq run (>100Gb) cannot be handled by existing quality control software like TileQC, SolexaQA, and HTQC [1–3] because of memory and/or computational errors. Second, SUGAR generates high resolution heatmaps showing spatial distributions of base quality, read density, and mapping quality scores on the flowcell. These heatmaps visualize low-quality spots of the flowcell that may be affected by technical errors due to air bubbles or other factors, notifying users to improve conditions of sequencing experiments. Third, not only to provide a quality assessment report [4], SUGAR also removes the low-quality read data or changes the low-quality nucleotide calls to N bases that were sequenced in the error-affected spots of the flowcell, by both manual operation and automated analysis using GUI guides. Such data cleaning enabled by the SUGAR would improve quality of sequence read mapping and downstream analyses.

The SUGAR successfully generated high resolution heatmaps from a human genome sequencing data (Figure 1; 10×10 resolution). We conducted this performance test using a HiSeq2000 data of HapMap individual NA12877 (ERA172924; read length of 100 bp and read coverage of 60× on average). The Fastq sequence reads were mapped to reference genome GRCh37 by the BWA [10]. The obtained 130GB BAM file was analyzed by SUGAR with computing environment with the Intel Xeon CPU E5-2640 processors (2.50 GHz). The CPU time and maximum memory size were less than 5 hours and 250 MB, respectively. An analysis of the corresponding Fastq file is much faster, generally finishes within an hour. Benchmark test was omitted in this study, because the existing tile-based quality control tools were not capable of processing whole HiSeq data using the computing environment described above.

The resultant heatmaps and curve chart-based data cleaning appear to contribute to improving both sequencing procedure and resultant data quality. First, the heatmaps indicate technical errors possibly arose from air bubbles (Figure 1A) and flowcell crack or primary data-processing error (Figure 1B). Such information would be useful to improve NGS machine conditions and sequencing workflows. Second, an average MapQ score of a HapMap individual NA12878 data (SRP021027; 100 bp data by the HiSeq2500; mapped by the Bowtie2 [11]) was increased from 29.5 to 30.1 after an automatic subtile deletions by applying the average QV threshold of 22 (>70% of data remains; Figure 1E). The discarded read data showed an average MapQ of 25.0. Although a conventional read quality-based data filtering provides improved MapQ score (e.g., 35.1 when the lower 30% reads were discarded from the same data set), a combination of such read-based filtering with the tile-based data cleaning of SUGAR provides higher MapQ score (e.g., 35.3 when the tile-based and read-based filtering were applied to the same data set). These results imply that low-quality subtiles may reduce an overall quality of the results of NGS data analyses, particularly in the case of low quality data sets, and thus the subtile-based data cleaning potentially has significant consequences. The detailed data evaluation and cleaning enabled by the SUGAR would reduce technical problems in sequence read mapping, improving subsequent analyses such as somatic mutation identification that require high-quality sequence data and mapping results. Taken together, we believe that the SUGAR, a GUI-based user-friendly tool, would contribute to controlling quality and reliability of upcoming high-throughput omics studies.

We developed a platform-independent java software SUGAR (subtile-based GUI-assisted refiner) to conduct quality evaluation and the quality-based cleaning of full HiSeq and MiSeq data through visualizing data quality distribution on the flowcell. The analysis can be done with user-friendly, GUI-based operations with relatively quick information processing and low memory space requirements (up to 250 MB for one data set). SUGAR will be especially useful to control the quality of variant calls of a sample with technical errors in sequencing procedures.

Project name: SUGAR.

Project home page: https://github.com/biomedinfo/sugar.

Operating system(s): Platform independent.

Programming language: Java.

Other requirements: Java version 6 or later.

License: GNU GPL version 3 or later.