Runcer-Necromancer: a method to rescue data from an interrupted run on MGISEQ-2000

Introduction

At the end of 2017, Chinese company MGI Tech presented the MGISEQ-2000 sequencing platform¹, promoting it as a device for large and medium scale genome sequencing. MGISEQ is specific in harnessing cPAS sequencing technology and using nanoballs (DNB) generated from circular molecules of DNA library by rolling circle replication². MGISEQ is compatible with a wide range of reagents for sequencing in SE50, SE100, SE400, PE100, PE150, and PE200 modes. MGISEQ-2000 provides the quality of sequencing comparable with that of the Illumina platform^3–6.

The first MGISEQ-2000 sequencer in Russia was installed in our lab (Center for Precision Genome Editing and Genetic Technologies for Biomedicine, Pirogov Medical University) in February 2019, and we run it once a week in the paired-end 150 mode (PE150). According to our experience, one PE150 run usually takes 68 hours if one flow cell is used at a time. During one of these runs, about 23.00 on Saturday, there was a failure of the Moscow power grid leading to a 50-minute blackout of a whole district including Pirogov Medical University. UPS battery storage was sufficient only for 20 extra minutes, then the sequencer turned off until the power was restored. Therefore, the device with loaded reagents remained in sleep mode for 35 hours until Monday. Before the instrument was switched off, it had performed 138 full cycles of forward read sequencing (run 27). The specific feature of the MGISEQ-2000 sequencing program is that it reads a barcode at the end of a run after it completes sequencing of forward and reverse reads. So, to a first approximation, the data obtained could not be demultiplexed as information on the barcodes was absent.

According to the MGI Tech⁷ recommendations, after consulting with the MGI Tech service engineers, we were advised to dispose of the current tank with reagents as well as the flow cell and run the samples using new reagents. In the first place, it is linked to the high sensitivity of the MDA reagent to storage at +4°C as it loses its activity very quickly. We decided to continue the run using the reagents that had been loaded for the weekend and try to restore the data. Finally, we managed to rescue the data using the software from ZebraCall⁸ and our own script on C++, which is reported here https://github.com/genomecenter/runcer-necromancer.

Methods

Sequencing

We prepared 3 pools of circularized libraries following the standard MGI Tech protocol⁹. Then we synthesized DNB, loaded a flow cell using the MGIDL-200H manual loader, prepared a sequencing cartridge from the MGISEQ-2000RS High-throughput Sequencing Set with User manual version: A2, and started sequencing on A-side in PE150 mode. Run 27 was aborted at the 139th cycle of the read-1 sequencing phase. After 35 h, we restarted the run (run 27_2) using the same sequencing cartridge and flow cell in a custom mode with the following parameters: read 1 for 12 cycles, read 2 for 151 cycles, Start phase: Sequencing (Figure 1). For summary reports generated by MGISEQ-2000 for lane 1 of runs 27 and 27_2, see Extended data: File S1, S2.

Figure 1. The screenshot of MGISEQ software in a custom mode with the settings used for restarting the run.

client.exe D:\Result\workspace\run_name\L01 139 6 72 -B C:\ZebraCallV2\empty_barcode.txt -N run_name -U 1 -F

Fastq merging

MGISEQ-2000 employs a patterned flow cell, so each DNB in a cell has unique coordinates at X and Y axes which do not depend on flow cell localization in a device and are not changed if the flow cell is displaced. When the power of the sequencer was off, the vacuum pump was switched off as well. The coordinates of each read were saved in a header of a .fastq file (Figure 2). This allowed us to integrate the data on forward reads obtained before and after the instrument was off.

Figure 2. The structure of a read header in an MGISEQ-2000 .fastq file.

As a read number being used for forward and reverse reading is unique, we managed to combine the 138-nucleotide sequences obtained during the first run with the nucleotide sequences obtained during the second run based on the information on F.O.V Column, F.O.V Row, and read numbers. To achieve this, we created a C++ script, which can be accessed at GitHub https://github.com/genomecenter/runcer-necromancer. The instruction for script running can be found below and in the file README.md in the repository.

Results

The most important parameter for sequencing quality is the ratio of the data with the quality level of no less than Q30. The Q30 value and other quality metrics did not decrease dramatically in spite of a 35-hour stand by (Figure 3, Table 1).

Figure 3.

Q30 histograms for runs 27 (А) and 27_2 (B). The X-axis represents the number of sequencing cycles, the Y-axis represents the ratio of the data with the quality no less than Q30 (%). Blue arrows in histogram B indicate the cycles which reverse reading and barcode reading started.

This implies that storing a loaded cartridge for 35 h leads to its decline, however, it can be still used for sequencing.

To check if merging reads from different runs was correct, we compared 7 samples of whole-exome sequencing from runs 27 and 27_2 with the data from the same samples obtained in run 22. We used the distribution of the size of an insert between left and right reads and the ratio of reads having the insert size exceeding 1000 nucleotides as control metrics. If read merging had been performed with errors, the portion of the reads mapped to various genome regions would have significantly increased. See Figure 4 for the distribution of an insert size for sample LWX777 from the control group.

Figure 4.

The distribution of inserts for sample LWX777 from runs 22 (A) and 27+27_2 (B). The X-axis represents the number of reads, the Y-axis represents an insert size. The diagrams were obtained using Picard CollectIsertSizeMetrics v2.22.4.

The ratio of reads from sample LWX777 with the insert size exceeding 1000 nucleotides was 0.003% in case of the data combined from the different runs, while it was 0.005% in case of the previous sequencing without read integration. The obtained data imply that read merging was correct.

Conclusion

It is possible to use a sequence cartridge after 35-hour storage at +4°C, although the quality of the obtained data is reduced.

We would like to point out that the method we propose can be applied in exceptional cases at your own risk and always followed by a quality check of the obtained data. As the procedure described above is not a complete experiment with replication and control samples, we cannot guarantee the quality of the data after merging reads of sequencing run under different problematic conditions. Therefore, we recommend that researchers in such situations adequately assess the conditions: how long a sequencing run was interrupted (hours or days), how the temperature and humidity in the laboratory room and inside the device changed, at which stage of sequencing (reading forward or reverse read, MDA reaction, reading a barcode) the power was turned off, etc. It is also desirable to use an in-lab reference sample in each run of the instrument to assess data quality and batch-effect. The described aborted run had a DNA library sample that we had previously sequenced under normal conditions, so we were able to assess the quality of the data obtained after merging reads in terms of GC-content, Ti/Tv and het/hom ratio, coverage statistics, variant calling.

Merging sequencing data can be successfully performed if the information about the localisation in flow cells is saved in a read header. The researcher must then compare the data quality on their own reference samples to decide whether to use the data from the aborted run.

Data availability

Software availability

Script available from: https://github.com/genomecenter/runcer-necromancer

Archived script as at time of publication: http://doi.org/10.5281/zenodo.4340350¹⁰.

License: MIT