`--barcoded` option fails as barcodes in latest Dorado summary files are listed in column `alias`

[m-jahn]

Hi!
The barcoded option does not work for summary text files generated with the latest Dorado basecaller v2.1.1.

The barcodes in the summary files are located in column alias, while column barcode is not present and barcode_arrangement just contains "unclassified". Here is an example of recent Dorado summary output. Note that it has more barcode related stats than it used to have a short while ago (v1.x).

input_filename	batch_id	parent_read_id	read_id	run_id	channel	mux	minknow_events	start_time	duration	passes_filtering	template_start	num_events_template	template_duration	sequence_length_template	mean_qscore_template	pore_type	experiment_id	sample_id	end_reason	alias	type	barcode_arrangement	barcode_kit	barcode_variant	barcode_score	barcode_front_score	barcode_front_foundseq_length	barcode_front_begin_index	barcode_rear_score	barcode_rear_foundseq_length	barcode_rear_end_index
FBC12581_83933ab2_794bc61c_0.pod5	0	acaf80f8-f984-4e39-8548-514fee8cdadc	acaf80f8-f984-4e39-8548-514fee8cdadc	83933ab2-043e-41c9-9be9-30beac509469	107	1	700	296.867004	1.131400	TRUE	296.869019	941	1.129400	408	23.301178	not_set	250319_23s-rRNA-deg_2	250319_23s-rRNA-deg_2	signal_positive	barcode07	na	unclassified	SQK-PCB114-24	var1	1.000000	1.000000	52	49	0.888889	49	406
FBC12581_83933ab2_794bc61c_0.pod5	0	aa71b5f3-0996-4f54-89d0-44d8c3fe7c4d	aa71b5f3-0996-4f54-89d0-44d8c3fe7c4d	83933ab2-043e-41c9-9be9-30beac509469	448	1	899	296.265991	1.546600	TRUE	296.268005	1287	1.544600	542	27.167379	not_set	250319_23s-rRNA-deg_2	250319_23s-rRNA-deg_2	signal_positive	barcode07	na	unclassified	SQK-PCB114-24	var2	0.923077	0.923077	55	42	0.833333	46	539
FBC12581_83933ab2_794bc61c_0.pod5	0	e184d049-e495-45dd-b774-790077fe5d2a	e184d049-e495-45dd-b774-790077fe5d2a	83933ab2-043e-41c9-9be9-30beac509469	11	1	1183	292.941010	1.911600	TRUE	292.943024	1591	1.909600	732	16.290915	not_set	250319_23s-rRNA-deg_2	250319_23s-rRNA-deg_2	signal_positive	barcode05	na	unclassified	SQK-PCB114-24	var1	0.974359	0.974359	52	44	0.777778	51	731

I can not say if this is always the case for all Nanopore data, so it might make sense if somebody can confirm this behavior.

[wdecoster]

Thanks for letting me know, I will investigate :-)

[wdecoster]

Thanks for the detailed report and the example file, that made this straightforward to track down.

The root cause is upstream in dorado rather than in the summary format. Dorado's summary writer sets both columns from the same barcoding result, with alias falling back to barcode_arrangement rather than the other way around:

barcode_arrangement = data.barcoding_result->normalized_barcode_name;
alias = data.barcoding_result->alias.empty() ? barcode_arrangement
                                             : data.barcoding_result->alias;

In 2.1.0 and 2.1.1 normalized_barcode_name was not being restored, so barcode_arrangement came out as unclassified for every read while alias kept the real barcode, exactly as in your file. It is fixed in dorado 2.1.2:

• 5f1bd6aa - Fix Dorado summary barcode_arrangement column always being set as unclassified.

So barcode_arrangement is still the right column to read, and this is not a change to the 26.01 output specification. To answer your question about whether it is always the case: no, only summary files written by dorado 2.1.0 and 2.1.1 are affected. Re-running dorado summary on the same BAM with 2.1.2 regenerates a correct file, no re-basecalling needed.

That said, summary files written by the affected versions are already out there and the failure was silent (NanoPlot happily produced a single unclassified_* report set rather than erroring out), so I have added a fallback. nanoget 1.19.6 and NanoPlot 1.47.3 are on PyPI now:

pip install -U nanoplot

When barcode_arrangement is unclassified for every read while alias holds real barcodes, nanoget now uses alias and prints a warning naming the dorado bug. The two cases that could be confused for this are both left alone: a genuinely unclassified run has unclassified in alias too, and a run with a sample sheet has real barcodes in barcode_arrangement, which stays authoritative.

On your file shape this now gives one report set per barcode again:

General summary:  barcode01  barcode02  barcode03  barcode04
Mean read length:   2,408.1    2,337.6    2,660.4    2,349.1

While in there I also fixed a related bug: nanoget renamed the summary columns by position, but read_csv returns columns in file order regardless of the order they are requested in, so any summary file not listing them in the expected order had every column silently mislabelled. Also fixed in 1.19.6.

[m-jahn]

Thanks so much for getting back so quickly and sorting this out thoroughly.
We will update Dorado to the latest version.