What are “pass-filter reads” and how do I calculate “pass-filter rate”?

This article applies to Cloudbreak^TM and Cloudbreak Freestyle^TM sequencing runs.

Introduction

Demultiplexing an AVITI™ run includes filtering of low-quality reads. By default, Bases2Fastq uses the first 15 cycles of each read for filtering, and this can be modified using the filter-mask flag in Bases2Fastq (see the Online Help for details).

If a sequencing run has lower density than expected, calculating how many reads pass the filtering criteria can provide valuable insight. A low proportion of pass-filter reads can indicate an issue with a sequencing run such as overloading, or an issue with a specific library or sample. The following terms will be useful for understanding these calculations:

• Pass-filter reads (PF reads): the total number of reads that have met the default filtering threshold in Bases2Fastq and are included in the final demultiplexing output.
• Fail-filter reads: reads that fail to meet the quality filtering threshold and are not included in the final demultiplexing output.
• Pass-filter rate (PF rate): the proportion of reads that pass filtering criteria, expressed as a percentage of the total reads. Generally, a standard sequencing run is expected to have a PF rate of ~80-85%, and an Expert Mode HD run is expected to have a PF rate of at least 70%. PF rates are not guaranteed.

Investigating PF Rate by Altering Demultiplexing Settings

Filter mask settings can be changed to investigate the PF rate for individual samples in a sequencing run. When conducting this kind of troubleshooting, it is recommended to demultiplex with no more than a few tiles from a run to reduce computational time. To assess PF rate, demultiplex the same tile(s) with both of the following settings:

1. Standard (default) settings: no changes except to number of tiles analyzed to match number analyzed without filtering.
2. No filtering: set the Bases2Fastq --filter-mask flag to R1:N*-R2:N*.

Instructions and examples for altering the filter-mask flag and selecting tiles for demultiplexing are included in the Optional Arguments section of the Bases2Fastq documentation. Keep all settings besides the filter-mask identical between the two demultiplexing runs so the results can be directly compared.

After demultiplexing, look for samples that are being filtered at a higher rate than others by comparing the IndexAssignment.csv files from each demultiplex output directory (one from the default demux and one from the no-filtering demux). The PF rate can be calculated for each sample by dividing the number of assigned reads from the default demux for that sample by the number of assigned reads with the no-filtering demux for the same sample. Comparing by assay or library type can make it easier to evaluate whether a particular type of library is being filtered more heavily than the others, as in Figure 1A below. Comparisons of library PF rate with PhiX PF rate from the same sequencing run is also possible. A low PF rate for PhiX may indicate an issue with the sequencing run, though PhiX can be impacted by other polonies on the flow cell surface.

In Figure 1, the PF rate has been calculated in panel A for four different assays from a single pool (each assay contains multiple samples, pooling was not equimolar). While PF rate is expected to vary a little by assay, Assay 3 stands out with a much lower PF rate (graphically represented in Figure 1B).

If a sample or library issue is suspected, the associated fastq files generated by the unfiltered demultiplex can be run through quality analysis tools like FastQC to check sequence length, quality, presence of adapters, and the expected profile based on the assay type. In Figure 1C, the FastQC report shows the samples from Assay 3 have significantly lower quality in Read 2, which is causing the heavy filtering of reads from that assay during demultiplexing.