HERRO’s correction results on ONT data

[majssssa]

I have a question regarding HERRO’s correction results on ONT data.

There is a very strange phenomenon: after HERRO correction, the data size decreased from 42 GB to 8 GB, yet the error rate does not seem to have improved substantially. The ONT data were obtained from the NCBI database, uploaded in 2023, and were likely sequenced using an R9 flow cell.

[M::ha_hist_line] 1: ****************************************************************************************************> 82053373
[M::ha_hist_line] 2: ****************************************************************************************************> 3578831
[M::ha_hist_line] 3: ****************************************************************************************************> 1224180
[M::ha_hist_line] 4: ****************************************************************************************************> 620384
[M::ha_hist_line] 5: ****************************************************************************************************> 379530
[M::ha_hist_line] 6: ****************************************************************************************************> 260118
[M::ha_hist_line] 7: ****************************************************************************************************> 192730
[M::ha_hist_line] 8: ****************************************************************************************************> 150846
[M::ha_hist_line] 9: ****************************************************************************************************> 123015
[M::ha_hist_line] 10: ****************************************************************************************************> 104784
[M::ha_hist_line] 11: ****************************************************************************************************> 91514
[M::ha_hist_line] 12: ****************************************************************************************************> 82277
[M::ha_hist_line] 13: ****************************************************************************************************> 74765
[M::ha_hist_line] 14: ****************************************************************************************************> 69698
[M::ha_hist_line] 15: ****************************************************************************************************> 65377
[M::ha_hist_line] 16: ****************************************************************************************************> 61760
[M::ha_hist_line] 17: ****************************************************************************************************> 59388
[M::ha_hist_line] 18: ****************************************************************************************************> 57435
[M::ha_hist_line] 19: ****************************************************************************************************> 55908
[M::ha_hist_line] 20: ****************************************************************************************************> 54394
[M::ha_hist_line] 21: ****************************************************************************************************> 53750
[M::ha_hist_line] 22: ****************************************************************************************************> 52290
[M::ha_hist_line] 23: ****************************************************************************************************> 51823
[M::ha_hist_line] 24: **************************************************************************************************** 50841
[M::ha_hist_line] 25: **************************************************************************************************** 50747
[M::ha_hist_line] 26: *************************************************************************************************** 50230
[M::ha_hist_line] 27: ************************************************************************************************* 49648
[M::ha_hist_line] 28: ************************************************************************************************* 49349
[M::ha_hist_line] 29: ************************************************************************************************* 49364
[M::ha_hist_line] 30: ************************************************************************************************* 49291
[M::ha_hist_line] 31: ************************************************************************************************* 49419
[M::ha_hist_line] 32: ************************************************************************************************ 49131
[M::ha_hist_line] 33: ************************************************************************************************* 49183
[M::ha_hist_line] 34: ************************************************************************************************ 48750
[M::ha_hist_line] 35: ************************************************************************************************* 49265
[M::ha_hist_line] 36: ************************************************************************************************** 49729
[M::ha_hist_line] 37: ************************************************************************************************* 49598
[M::ha_hist_line] 38: ************************************************************************************************* 49653
[M::ha_hist_line] 39: *************************************************************************************************** 50250
[M::ha_hist_line] 40: *************************************************************************************************** 50645
[M::ha_hist_line] 41: *************************************************************************************************** 50611
[M::ha_hist_line] 42: **************************************************************************************************** 50853
[M::ha_hist_line] 43: *************************************************************************************************** 50545
[M::ha_hist_line] 44: ************************************************************************************************** 50096
[M::ha_hist_line] 45: *************************************************************************************************** 50606
[M::ha_hist_line] 46: ************************************************************************************************** 50157
[M::ha_hist_line] 47: **************************************************************************************************** 50961
[M::ha_hist_line] 48: *************************************************************************************************** 50364
[M::ha_hist_line] 49: *************************************************************************************************** 50205
[M::ha_hist_line] 50: ************************************************************************************************** 50128
[M::ha_hist_line] 51: ************************************************************************************************* 49522
[M::ha_hist_line] 52: *********************************************************************************************** 48160
[M::ha_hist_line] 53: ********************************************************************************************* 47436
[M::ha_hist_line] 54: ******************************************************************************************** 46679
[M::ha_hist_line] 55: ****************************************************************************************** 45961
[M::ha_hist_line] 56: *************************************************************************************** 44117
[M::ha_hist_line] 57: ************************************************************************************* 43121
[M::ha_hist_line] 58: ********************************************************************************** 41984
[M::ha_hist_line] 59: ******************************************************************************* 40088
[M::ha_hist_line] 60: **************************************************************************** 38548
[M::ha_hist_line] 61: ************************************************************************* 37077
[M::ha_hist_line] 62: ********************************************************************* 35289
[M::ha_hist_line] 63: ***************************************************************** 33336
[M::ha_hist_line] 64: ************************************************************** 31659
[M::ha_hist_line] 65: ********************************************************** 29743
[M::ha_hist_line] 66: ****************************************************** 27704
[M::ha_hist_line] 67: ************************************************** 25416
[M::ha_hist_line] 68: *********************************************** 23957
[M::ha_hist_line] 69: ******************************************* 21766
[M::ha_hist_line] 70: *************************************** 19743
[M::ha_hist_line] 71: ********************************** 17334
[M::ha_hist_line] 72: ******************************* 15664
[M::ha_hist_line] 73: *************************** 13845
[M::ha_hist_line] 74: ************************ 12214
[M::ha_hist_line] 75: ******************** 10415
[M::ha_hist_line] 76: ****************** 9167
[M::ha_hist_line] 77: **************** 7949
[M::ha_hist_line] 78: ************* 6546
[M::ha_hist_line] 79: *********** 5552
[M::ha_hist_line] 80: ********* 4716
[M::ha_hist_line] 81: ******* 3815
[M::ha_hist_line] 82: ****** 3290
[M::ha_hist_line] 83: ***** 2684
[M::ha_hist_line] 84: **** 2282
[M::ha_hist_line] 85: **** 1896
[M::ha_hist_line] 86: *** 1580
[M::ha_hist_line] 87: *** 1282
[M::ha_hist_line] 88: ** 1116
[M::ha_hist_line] 89: ** 866
[M::ha_hist_line] 90: * 691
[M::ha_hist_line] 91: * 623
[M::ha_hist_line] 92: * 476
[M::ha_hist_line] 93: * 441
[M::ha_hist_line] 94: * 351
[M::ha_hist_line] 95: * 300
[M::ha_hist_line] 96: * 293
[M::ha_hist_line] 97: * 275
[M::ha_hist_line] rest: ************* 6450

[dominikstanojevic]

Hi @majssssa, thanks for reaching out! A few thoughts and follow-up questions:

Data size reduction
The ~80% size drop is likely caused by two compounding factors:

1. The format conversion from FASTQ → FASTA drops quality scores, which alone should roughly halve the file size (42 GB → ~21 GB).
2. The remaining reduction is expected HERRO behaviour — reads with insufficient alignments (i.e. too few overlapping reads to correct against) get trimmed, split, or discarded entirely. This is more pronounced with noisy data where alignments may be sparse or unreliable.

Which HERRO model did you use?
This is likely the most important question. HERRO ships with separate models for R9 (experimental; haven't performed detailed benchmarking) and R10.4.1 data. If you used the R10 model on R9 reads, the correction quality would be significantly degraded. Could you confirm which model you passed to HERRO?

Regarding the histogram
Could you clarify how this histogram was generated? The [M::ha_hist_line] prefix looks like hifiasm output — if so, was it run on the raw reads? This would help us interpret whether the k-mer distribution reflects the input data quality or something introduced during correction. The large spike at depth 1 does suggest high noise in the data, which is consistent with R9 reads. For all-vs-all alignments for R9 data you will need to use minimap2, I highly doubt that hifiasm (with --dbg-ovec) would work for R9 data.

Once we know the model used and the source of the histogram, we should be able to narrow down the root cause.

[majssssa]

Thank you for the suggestions.

We used the HERRO R9 model for correction. The k-mer histogram was generated from hifiasm in the HERRO-corrected reads rather than the raw reads. For the all-vs-all overlap step during correction, we used minimap2 rather than hifiasm (--dbg-ovec).

Given these settings, it seems that neither model mismatch nor the overlap generation method is likely to explain the substantial reduction in data size. Do these results suggest that the underlying read quality may be the primary factor, or are there any additional diagnostics you would recommend to further investigate the issue?

One additional consideration is that these sequencing data were not generated by us but were downloaded from a public NCBI dataset. Therefore, we cannot independently verify the sequencing chemistry or flow cell version. Although the dataset was described as nanopore data, it is possible that the reads were generated using an older flow cell rather than R9 or R10 chemistry, which could have resulted in substantially higher raw error rates and consequently a greater degree of read filtering or correction during HERRO processing.

[dominikstanojevic]

Thanks for the detailed follow-up! Could you share the SRA accession number for the dataset? That would allow us to investigate further on our end.

[majssssa]

Thanks，SRA accession number is SRR23701665.

[txfc]

I used Herro to calibrate the ont data, but I found that the output can only be set to FASTA format, because hifiasm (ont) can only accept the input of FASTQ. Is it better to use Verkko to do the downstream assembly?