Questions about SQANTI3 vs. SQANTI-single-cell default parameters (min_cov and ratio_TSS) for single-cell long-read data

[lihua061777-ops]

I am using SQANTI3 to process long-read single‑cell RNA‑seq data and have noticed default parameter differences between SQANTI3 and SQANTI‑single‑cell. I’d appreciate your clarification.

--min_cov: In SQANTI3, the default is 3; in SQANTI‑single‑cell, it’s 1. Is this change intended to accommodate the sparser nature of single‑cell data? For my single‑cell dataset, can I simply set min_cov=1 in a custom JSON filter when using SQANTI3?

--ratio_TSS: In SQANTI‑single‑cell, there is a --ratio_TSS parameter with a default value of 2.0. Is this 2.0 threshold also applicable as a reference when running the standard SQANTI3 pipeline on single‑cell data? If so, should I enforce it by adding a rule for the ratio_TSS column in a filtering JSON file?

[CarlosBlancoGo]

Hi @lihua061777-ops , thank you for your question.

The important thing to understand is that these are not parameters that change how SQANTI3 performs its structural classification. min_cov and ratio_TSS are columns that SQANTI3 populates in the classification file using short-read data, when you provide the relevant input (--coverage for junction support, --SR_bam or --short_reads for TSS coverage). You can then use those columns as criteria in a filter JSON as you were commenting if you wish, but that is a decision given to the filter module for post-QC filtering.

In SQANTI-single-cell, these parameters are not yet implemented as filtering decisions, as the filter module is still under development. However, they are implemented as CLI options to give to the QC module as a way to establish a threshold for short-read QC validation. SQANTI-single-cell creates a cell-level classification file called the cell summary, containing values for different metrics for each cell barcode. Typically these metrics are normalized, showing them as the proportion of reads/transcripts in the cell barcode that satisfy a given requirement (e.g. being a FSM, having a length above 2000bp, etc.). These two metrics follow the same logic: we needed to establish a threshold so that the feature could represent something like "proportion of reads/transcripts in the cell whose junctions are validated by orthogonal short reads" and "proportion of reads/transcripts in the cell whose TSS is validated by orthogonal short reads."

The specific reasons for those default thresholds:

• For min_cov, in the single-cell context the short reads used for junction validation cannot come from the same 10x Chromium experiment. Because 10x for short-reads captures only the 3' end of the transcript, most splice junctions across the gene body are never covered by those reads. The way SQANTI-sc handles this is that the junction validation requires a separate bulk RNA-seq dataset (ideally from the same tissue and condition). Since this bulk data comes from a different experiment, it may not represent the full diversity of cell-type-specific splice junctions present in your single-cell sample, particularly for novel or rare junctions expressed only in a small subset of cells. We therefore considered that a lower threshold was more appropriate to avoid discarding genuine cell-type-specific junctions simply due to low bulk short-read support.
• For ratio_TSS, the same limitation applies: 10x short reads are 3'-biased (if using 3' design kits) and do not cover the TSS region, so TSS validation also requires bulk short reads from a separate experiment. A ratio above 1 suggests the TSS is real, and 2.0 means the signal is at least twice the background. We chose this threshold as a balance between filtering out isoforms with no clear TSS signal (ratio close to 1, indistinguishable from background) and retaining genuine alternative TSSs, which can be biologically possible, but may show only modest short-read enrichment if the expressing cell population that expresses them is small.

I'm curious about how were you planning to tackle the filtering of your isoforms using short-read data at single-cell level.