BCL Convert Performance Gains

[thefabnab]

Hey Folks,

It seems Illumina has recently released BCL Convert as a long term replacement for Bcl2Fastq. Has anyone taken this tool out for a spin yet? Any performance gains/recommendations they for implementation?

I'm seeing that they've released RPMs for CentOS. Has anyone had any issues porting that to Ubuntu?

[GenoMax]

It appears to be about 2x faster compared to bcl2fastq on identical datasets. It is producing different folder structure for result/log files (no more html reports) but plain csv so they should be easily parseable.

[kcchan]

bcl-convert installs just fine in Ubuntu. The process is the same as installing bcl2fastq2

About 90% of the features from bcl2fastq were missing until the most recent release, but now it's mostly feature complete. Definitely read through the user manual and FAQ before starting as almost all of the settings have changed. https://support.illumina.com/bulleti...l-convert.html

Something I personally find rather annoying is that many of the settings have moved into the SampleSheet.csv file. So now you have to edit the csv file for every little thing such as adjusting the read masking and mismatch settings, but I guess some people prefer that.

I haven't seen the 2x speed improvement, but with the NextSeq data I've tested the improvement is around 25%.

[j13kagrf]

More information for others upgrading/migrating to BCL Convert

As well as requiring most settings to be specified in the sample sheet file, it's worth noting that the output reports have also changed with BCL Convert, as this can affect pipeline steps that consume those reports.

For example, ConversionStats.xml, which contained Raw cluster count, read number, yield Q30, yield etc. at the tile level, is no longer output to a Stats/ subfolder. In fact that directory has been dropped.

Instead, BCL Convert outputs some metrics in new CSV files to the Reports/ output subfolder. Some of the metrics from ConversionStats.xml are output to the new Quality_Metrics.csv report, but are aggregated at the lane/barcode/read grouping rather than the tile level.