Summary.txt with only zeros/ rMATS_result_FDR.txt not created - NO ERROR MESSAGE

[mlbonatelli]

Description of the bug

Hi! I am using nf-core/rnasplice from .BAM files to evaluate the alternative splicing in my samples. I previously ran nf-core/rnaseq with star, and I got the .BAM files from there. I am particularly interested in the output of rmats.

The first time I ran, I used three biological replicates for the control and three for each treatment. There, the pipeline worked fine, but I got only a few splicing events that where statistically significant. So I excluded one of my replicates in the control sample, because it seemed a bit odd.

I ran again the pipeline in the same way, but I used a different input list, not considering the the control sample I didnt want to. The pipeline finished without errors, but Summary.txt contained only zeros and rMATS_result_FDR.txt was not created. Splicing events were detected (see below 2025-07-02-18_02_42_078011_read_outcomes_by_bam.txt). But for some reason, the rmats_post step ran particularly fast (only 15 s - see nextflow-log.txt and execution_trace_2025-07-02_11-33-17.txt) and did not result in the proper output.

I checked many reasons for the Summary.txt to have only zero values on the rmats github, but I could not find what was going on. Since it ran before without any issues with the triplicate samples in control, but now it does not with duplicates.

I would appreciate any help. If you need more info, please let me know.

Command used and terminal output

nextflow run \
    nf-core/rnasplice \
    --input $OUTDIR/alignments.csv \
    --contrasts $OUTDIR/contrasts.csv \
    --fasta $GENOME/Arabidopsis_thaliana.TAIR10.dna.toplevel.fa.gz \
    --gtf $GENOME/Arabidopsis_thaliana.TAIR10.60.gtf.gz \
    --outdir $OUTDIR \
    --source genome_bam \
    --sashimi_plot false \
    -profile singularity

Relevant files

2025-07-02-18_02_42_078011_read_outcomes_by_bam.txt
execution_trace_2025-07-02_11-33-17.txt
nextflow-log.txt

System information

• Nextflow version: 25.04.2
• Hardware: HPC
• Executor: slurm
• Container engine: singularity
• OS: Linux
• Version of nf-core/rnasplice: v1.0.4-g1d0494a

[mlbonatelli]

I have not discovered how to fix this issue, but just as an update: I ran rmats on the same samples and I got a result that makes sense with our biological question. Just in case anyone runs into the same issue as I did.

[pablolopeznie]

Hi @mlbonatelli,

I am currently facing this exact problem. I did a try run with 3 replicates control-treatment and got rMATS to work properly. Afterwards, I used the same pipeline with a bigger dataset and encountered this trouble. From what I have investigated on, other users have reported it as an issue with some R packages in older versions of rMATS.

If I have understood it right, you ran rMATS independently of nf-core/rnasplice pipeline and you got coherent results?

Thanks in advance,
Pablo

[mlbonatelli]

Hi @pablolopeznie,

Yes, you are correct. I created a conda environment for rmats/4.3.0, and was able to run the analysis without any problem. It was quite straightforward :)

[pablolopeznie]

Hi again!

I finally found the reason of failure after some investigation.

By default, rnasplice sets --rmats_paired_stats argument to true with no further checking.

According to rMATS official documentation, this parameter can only be used when paired data are introduced: that is to say, when the number of samples in treatment condition is the same that in control condition, so that analysis is made in pairs (a1,b1,c1 vs a2,b2,c2). In my case, I was trying to make a contrast between a cohort of 10 cases vs. 3 controls, provoking the following error:

Loading data...
Error in load.data(my.data) : 
  Error with data: some data are not matched pairs.
Calls: <Anonymous> -> clean.data -> load.data
Execution halted

@mlbonatelli I created a dedicated conda environment as you proposed earlier to avoid running the entire pipeline again, thanks! :)

[piplus2]

I think that the default behaviour should be better documented. I'll add that to the dev branch.

[piplus2]

I have not discovered how to fix this issue, but just as an update: I ran rmats on the same samples and I got a result that makes sense with our biological question. Just in case anyone runs into the same issue as I did.

can you please show the command that ran correctly?

[piplus2]

Also, by default rnasplice sets rmats_paired_stats = true in nextflow.config that, as said by @pablolopeznie requires paired columns. You can run the pipeline with --rmats_paired_stats false to avoid that behaviour.

[piplus2]

Removing the label bug, as it is about improving the documentation.

[mlbonatelli]

Hey @piplus2, I've used rmats 4.3.0, and here is the command:

rmats.py --b1 $DIR/col.txt --b2 $DIR/"$input_file".txt
--gtf $GENOME/Arabidopsis_thaliana.TAIR10.60.gtf
-t paired --readLength 150 --variable-read-length --nthread $SLURM_CPUS_PER_TASK
--od $DIR/output_"$input_file" --tmp $DIR/tmp_output_"$input_file"

DIR= home directory
GENOME= location genome
col.txt = file with realpath from the bam files of my control samples
"$input_file".txt = file with realpath from the bam files of my treatment samples

[piplus2]

Thanks! That confirms what you said, the issue is the rmats_paired_stats param which is true by default. I've just pushed a change in the doc to better explain the effect of that param. I hope it helps.