Running with Parse Biosciences single-cell long-read data using pre-called barcodes

[evanwangx]

Hello,

I have ParseBio + Nanopore bam files with pre-called barcodes from the Parse pipeline. Since Parse isn't explicitly listed in your single-cell mode documentation, I'm using --barcoded_bam with --barcode_tag CB and --umi_tag PN to bypass barcode calling. However, I'm unsure which --mode to specify for a non-native platform (for UMI deduplication). Is there a way to specify UMI length independently of the mode? (I don't think custom_sc is correct here since barcode calling has already been performed).

Running IsoQuant version 3.12.2

Thanks!

--

Edit:

I ended up trying with custom_sc without specifying a MDF and PCR filtering occurred according to the logs:

2026-04-16 05:52:33,275 - INFO - Filtering PCR duplicates with edit distance 4

The Parse UMIs are 10bp long. Is edit distance of 4 the correct threshold? And do you have any recommendations on parameters? Deduplication stats:

2026-04-16 06:01:01,181 - INFO -   Unique gene-barcodes pairs: 7386671
2026-04-16 06:01:01,181 - INFO -   Total reads saved: 10416079
2026-04-16 06:01:01,181 - INFO -   Spliced reads saved: 2995860
2026-04-16 06:01:01,181 - INFO -   Total assignments processed: 15403919
2026-04-16 06:01:01,181 - INFO -   Assigned to any gene: 13593108
2026-04-16 06:01:01,181 - INFO -   Assigned to any gene and barcoded: 13238374
2026-04-16 06:01:01,181 - INFO -   Spliced: 3562888
2026-04-16 06:01:01,181 - INFO -   Spliced and barcoded: 3412169
2026-04-16 06:01:01,181 - INFO -   Uniquely assigned: 13238374
2026-04-16 06:01:01,181 - INFO -   Uniquely assigned and barcoded: 13238374
2026-04-16 06:01:01,181 - INFO -   Uniquely assigned and spliced: 3412169
2026-04-16 06:01:01,181 - INFO -   Uniquely assigned and spliced and barcoded: 3412169 

[andrewprzh]

Dear @evanwangx

Unfortunately, there is no way to set UMI length separately. However, the only parameter it affects is indeed the edit distance threshold used for UMI deduplication.

4 seems ok, can be a bit too conservative. You can also set --mode tenX_v2 which assumes UMIs of length exactly 10bp. This (and in fact tenX_v3) will set the threshold to 3.

Best
Andrey

[evanwangx]

Hi Andrey,

Thanks for your response. I am processing Parse single-nuc RNA-seq + Nanopore data using the revised workflow:

Step 1: Bulk mode discovery using combined BAMs to generate a consensus transcript_models.gtf
Step 2: Per sample quantification in --mode tenX_v2 with --no_model_construction, --barcoded_bam, --barcode_tag CB and --umi_tag pN against the consensus GTF

I'm seeing a large drop between total assignments (~55M) and total assignments processed in UMI filtering (~19M). This is consistent for each sample. Only ~35% of assigned reads are retained. My data has low polyA detection (~3%) and short N50 (~800bp), which I suspect inflates ambiguous and inconsistent assignments.

Is this drop directly related to the low polyA detection and N50, or does it suggest something is misconfigured?
Which read assignment categories are excluded before UMI filtering in single-cell mode, and is this intentional?

Thanks!

[andrewprzh]

Dear @evanwangx

The pipeline looks ok. PolyA presence and read length should not affect reads usage directly.
The reads is used by UMI deduplication algorithms if it (1) has a barcode, (2) is assigned to a gene uniquely and consistently (not to an isoform, just the gene). In a read has unique inconsistent assignment (e.g. novel splice site) it will also be sued.

Could you send me the log and maybe some statistics about assignment types based on read_assignments? That will help to understand the process.

Best
Andrey

[evanwangx]

Hi Andrey,

I've attached one of the quantification logs and stats from a few samples.

Thanks so much.

o121519440_isoquant.log

isoquant_metrics_comparison.tsv

[andrewprzh]

Yes, this kind of drop is a bit unusual. If you have the possibility, could you run IsoQuant in bulk mode so we can investigate assignment types before UMI filtering?

[evanwangx]

Hi Andrey,

I've appended the bulk read assignment statistics to the bottom in the attached file. The samples were run against the same combined GTF and without any single cell flags.

Thanks

isoquant_metrics_comparison_bulk.tsv

[andrewprzh]

@evanwangx

You data contains huge amount of "noninformative" reads, meaning they do not overlap with the annotated genes.
Most likely, either there is a problem with the annotation, or there is something is off with library preparation.
Have you tried using some default annotations like Gencode?

Best
Andrey