SUPPA2 on single cell data generated on 10x and Pacbio platform

[leewqa]

Hi,

I was wondering if SUPPA2 can be used for aggregated (pseudobulk by cell cluster) single cell long-read sequencing data generated using droplet-based 10x platform. And if so, what would be the appropriate count matrix to use? Would it be counts per million (CPM) for each sample?

Thanks a lot!

Best,
Audrey

[EduEyras]

Hi Audrey Thanks for your message SUPPA is agnostic to the experimental approach used to obtain transcript expression for estimating splicing changes. We have only tested SUPPA using TPM values. For long-read sequencing data, if each read corresponds to a transcript copy (this is true e.g. for techniques where you only sequence from the 3' end, i.e. no internal priming), then you can consider TPM ~ CPM, and it should work. I hope this helps Eduardo

…

On Wed, 25 Mar 2026 at 20:35, auderl ***@***.***> wrote: *leewqa* created an issue (comprna/SUPPA#222) <#222> Hi, I was wondering if SUPPA2 can be used for aggregated (pseudobulk by cell cluster) single cell long-read sequencing data generated using droplet-based 10x platform. And if so, what would be the appropriate count matrix to use? Would it be counts per million (CPM) for each sample? Thanks a lot! Best, Audrey — Reply to this email directly, view it on GitHub <#222?email_source=notifications&email_token=ADCZKBZRKWKUND24Y66ZNB34SOR7LA5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NBRGMZTSOBQGM3TNJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADCZKBYR3PXCH4C545BYWOD4SOR7LAVCNFSM6AAAAACW6TQGCGVHI2DSMVQWIX3LMV43ASLTON2WKOZUGEZTGOJYGAZTONQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[leewqa]

Hi Eduardo,

I had ran SUPPA2 psiPerEvent and psiPerIsoform successfully. However, when I check the delta PSI, it seems like the direction of the psi event is different than expected.

For example, I had ran my diffSplice function using this line: -e ${INPUT_FILE_1} ${INPUT_FILE_2}
Should I be expecting delta psi in the output to be INPUT_FILE_1 - INPUT_FILE_2?
This is the result I got for 1 isoform from diffSplice on psiPerIsoform:
ENSG00000169228.14;ENST00000478234.5 -0.10348762 0.03746254

In this case, I am expecting this isoform "ENSG00000169228.14;ENST00000478234.5" to be lower in INPUT_FILE_1 vs INPUT_FILE_2 right? But when I manually calculated psi in the CPM file, INPUT_FILE_1 should have a higher CPM percentage than INPUT_FILE_2. Another tool also confirmed that INPUT_FILE_1 has a higher proportion of this isoform than INPUT_FILE_2.

Would you be able to advise? Am I interpreting the delta psi wrongly? Thanks a lot.

[EduEyras]

Hi Audrey The delta PSI is calculated suppa.py diffSplice ... --psi <Cond1.psi> <Cond2.psi> ... Where Event PSI difference (ΔPSI) between Cond1 and Cond2 is calculated as ΔPSI = PSI_2 - PSI_1. Since the PSI not only depends on the TPM of the isoform, but also on the overall TPM of all other isoforms in the gene, it is possible that the TPM is larger in the case where the PSI is smaller, i.e. the proportion of the total gene expression explained by the isoform is smaller. You mention CPM but the correct number to use is the TPM. It might be worth plotting the TPMs of the isoforms in the gene, and how much of the gene expression they explain to see what's happening. The TPM is normalised by the transcript length, so the values are comparable. If you use CPMs or any other number that is not normalised by the transcript length, the results might be confounded by the number of reads or the total gene expression. I hope this helps Eduardo

…

On Mon, 13 Apr 2026 at 21:44, auderl ***@***.***> wrote: *leewqa* left a comment (comprna/SUPPA#222) <#222 (comment)> Hi Eduardo, I had ran SUPPA2 psiPerEvent and psiPerIsoform successfully. However, when I check the delta PSI, it seems like the direction of the psi event is different than expected. For example, I had ran my diffSplice function using this line: -e ${INPUT_FILE_1} ${INPUT_FILE_2} Should I be expecting delta psi in the output to be INPUT_FILE_1 - INPUT_FILE_2? This is the result I got for 1 isoform from diffSplice on psiPerIsoform: ENSG00000169228.14;ENST00000478234.5 -0.10348762 0.03746254 In this case, I am expecting this isoform "ENSG00000169228.14;ENST00000478234.5" to be lower in INPUT_FILE_1 vs INPUT_FILE_2 right? But when I manually calculated psi in the CPM file, INPUT_FILE_1 should have a higher CPM percentage than INPUT_FILE_2. Another tool also confirmed that INPUT_FILE_1 has a higher proportion of this isoform than INPUT_FILE_2. Would you be able to advise? Am I interpreting the delta psi wrongly? Thanks a lot. — Reply to this email directly, view it on GitHub <#222 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADCZKBZXTJJFUJVPE6TU2Q34VTHJVAVCNFSM6AAAAACW6TQGCGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DEMZWGEYDSMZRGQ> . You are receiving this because you commented.Message ID: ***@***.***>

[ExctOcean]

Hi Eduardo,

Thank you for the previous clarifications. I am currently working on Nanopore single-cell cDNA Direct-seq (10x platform). Due to sample limitations, I have only one biological replicate per experimental condition.

I would like to perform differential alternative splicing (DAS) analysis across different conditions/cell types using SUPPA2 and would appreciate your advice on which of the following two strategies is more statistically sound or supported by SUPPA2:

Option 1: Direct Pseudo-bulk Comparison

Generate a single Pseudo-bulk TPM matrix for each cell type per condition (aggregating all cells within that cluster) and run SUPPA2 diffSplice directly. However, I am concerned that the lack of replicates might affect the empirical p-value calculation or the robustness of the ΔPSI.

Option 2: Pseudo-replicates via Subsampling

For each Condition x Cell-type group, I could artificially create three "pseudo-replicates" by randomly subsampling cells (stratified sampling) and generating their respective Pseudo-bulk TPM matrices. These would then be fed into SUPPA2 to provide the "replicates" required for the statistical test.

Questions:

Given that SUPPA2’s diffSplice uses the variability between replicates to estimate significance, does SUPPA2 support Option 1 (n=1 per group)? If so, how is the significance handled?

Is Option 2 (pseudo-replicates) a recommended workaround in the community for single-cell data when biological replicates are missing, or would this lead to an inflation of false positives?

Which approach would you suggest for Nanopore-based long-read data to achieve the most reliable DAS results?

Thanks again for your time and help!

Best regards,

ExctOcean

[EduEyras]

Hi ExctOcean Thanks for your message. Suppa cannot estimate significance from a single sample, so it will be better to use the second option. Pseudo-replicates are also used by other methods to provide a distribution for the estimates of inclusion or expression levels, so it should work well in this case I hope this helps cheers Eduardo

…

On Thu, 16 Apr 2026 at 23:10, ExctOcean ***@***.***> wrote: *ExctOcean* left a comment (comprna/SUPPA#222) <#222 (comment)> Hi Eduardo, Thank you for the previous clarifications. I am currently working on Nanopore single-cell cDNA Direct-seq (10x platform). Due to sample limitations, I have only one biological replicate per experimental condition. I would like to perform differential alternative splicing (DAS) analysis across different conditions/cell types using SUPPA2 and would appreciate your advice on which of the following two strategies is more statistically sound or supported by SUPPA2: Option 1: Direct Pseudo-bulk Comparison Generate a single Pseudo-bulk TPM matrix for each cell type per condition (aggregating all cells within that cluster) and run SUPPA2 diffSplice directly. However, I am concerned that the lack of replicates might affect the empirical p-value calculation or the robustness of the ΔPSI. Option 2: Pseudo-replicates via Subsampling For each Condition x Cell-type group, I could artificially create three "pseudo-replicates" by randomly subsampling cells (stratified sampling) and generating their respective Pseudo-bulk TPM matrices. These would then be fed into SUPPA2 to provide the "replicates" required for the statistical test. Questions: Given that SUPPA2’s diffSplice uses the variability between replicates to estimate significance, does SUPPA2 support Option 1 (n=1 per group)? If so, how is the significance handled? Is Option 2 (pseudo-replicates) a recommended workaround in the community for single-cell data when biological replicates are missing, or would this lead to an inflation of false positives? Which approach would you suggest for Nanopore-based long-read data to achieve the most reliable DAS results? Thanks again for your time and help! Best regards, ExctOcean — Reply to this email directly, view it on GitHub <#222 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADCZKB3MDPSAUOUAKFXKNTD4WDLUFAVCNFSM6AAAAACW6TQGCGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DENRQGMYDGOBUGQ> . You are receiving this because you commented.Message ID: ***@***.***>

[ExctOcean]

Hi ExctOcean

Thanks for your message.

Suppa cannot estimate significance from a single sample, so it will be
better to use the second option. Pseudo-replicates are also used by other
methods to provide a distribution for the estimates of inclusion or
expression levels, so it should work well in this case

I hope this helps

cheers

Eduardo
…

On Thu, 16 Apr 2026 at 23:10, ExctOcean @.***> wrote:
ExctOcean left a comment (comprna/SUPPA#222)
<#222 (comment)>

Hi Eduardo,

Thank you for the previous clarifications. I am currently working on
Nanopore single-cell cDNA Direct-seq (10x platform). Due to sample
limitations, I have only one biological replicate per experimental
condition.

I would like to perform differential alternative splicing (DAS) analysis
across different conditions/cell types using SUPPA2 and would appreciate
your advice on which of the following two strategies is more statistically
sound or supported by SUPPA2:

Option 1: Direct Pseudo-bulk Comparison

Generate a single Pseudo-bulk TPM matrix for each cell type per condition
(aggregating all cells within that cluster) and run SUPPA2 diffSplice
directly. However, I am concerned that the lack of replicates might affect
the empirical p-value calculation or the robustness of the ΔPSI.

Option 2: Pseudo-replicates via Subsampling

For each Condition x Cell-type group, I could artificially create three
"pseudo-replicates" by randomly subsampling cells (stratified sampling) and
generating their respective Pseudo-bulk TPM matrices. These would then be
fed into SUPPA2 to provide the "replicates" required for the statistical
test.

Questions:

Given that SUPPA2’s diffSplice uses the variability between replicates to
estimate significance, does SUPPA2 support Option 1 (n=1 per group)? If so,
how is the significance handled?

Is Option 2 (pseudo-replicates) a recommended workaround in the community
for single-cell data when biological replicates are missing, or would this
lead to an inflation of false positives?

Which approach would you suggest for Nanopore-based long-read data to
achieve the most reliable DAS results?

Thanks again for your time and help!

Best regards,

ExctOcean

—
Reply to this email directly, view it on GitHub
<#222 (comment)>, or
unsubscribe
https://github.com/notifications/unsubscribe-auth/ADCZKB3MDPSAUOUAKFXKNTD4WDLUFAVCNFSM6AAAAACW6TQGCGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DENRQGMYDGOBUGQ
.
You are receiving this because you commented.Message ID:
@.***>

Hi Eduardo,

Thank you very much for your prompt feedback and for clarifying the statistical requirements of SUPPA2.

Following your suggestion, I will implement the pseudo-replicate strategy in my research to provide the necessary distribution for significance estimation.

I truly appreciate your support and guidance on this!

Best regards,

ExctOcean