Isoform content differs when aligning to transcriptome reference compared to when aligning to genome (and post-processing with FLAIR)

[lgraciamaz]

I have read the paper (https://doi.org/10.1038/s41467-020-15171-6)
and the manual (https://flair.readthedocs.io/en/latest/) and I still have a question about

• correct
• collapse
  -quantify

I have aligned my ONT DRS reads with minimap2 using the command below:

minimap2 -ax splice -uf -k14 --secondary=no /mnt/f/Nanopore/GRCh38.primary_assembly.genome.fa FBC72805_pass_9f0f5c78_420f17b9_0.fastq.gz > C1_splicing_alignment_04-06-2026.sam

Then, I convert to bam, index, and convert to bed with the following commands:

samtools view -bS splicing_alignment_03-24-2026.sam | samtools sort -o splicing_alignment_03-24-2026.sorted.bam

samtools index splicing_alignment_03-24-2026.sorted.bam

bedtools bamtobed -bed12 -i splicing_alignment_03-24-2026.sorted.bam > reads_03-24-26.bed

Then, I proceed to correct, collapse, and quantify with the commands below:

flair correct -q reads_03-24-26.bed -f /mnt/f/Nanopore/gencode.v49.primary_assembly.annotation.gtf -o flair.corrected

flair collapse -g /mnt/f/Nanopore/GRCh38.primary_assembly.genome.fa --gtf /mnt/f/Nanopore/gencode.v49.primary_assembly.annotation.gtf -r FBC72805_pass_565e44b9_4ab34d15_0.fastq.gz -q flair.corrected_all_corrected.bed -o flair.isoformsv2

flair quantify -r {sample}_manifest.tsv -i {sample}_reads.fastq.gz -t 8 -o {sample}_flair.quantify

I then use IsoVis (https://isomix.org/isovis/) to visualize the isoform contents of my sequencing experiments.

However, when I perform alignment using the same minimap2 command but using gencode.v49.primary_assembly.annotation.gtf as reference, I get different isoform content than when I align to GRCh38.primary_assembly.genome.fa and then process with FLAIR. i.e., When I align to the transcriptome, I see significant coverage to transcript isoforms that are not present in the final quantify.csv file obtained from aligning to genome and then processing with the FLAIR modules mentioned above.

Is it possible that I am losing information when processing with FLAIR?

Thank you in advance for your assistance on this matter.

Kind regards,

Luis Gracia Mazuca

[cafelton]

Hi Luis,
Aligning to transcriptomes is a tricky business. Many reads align to transcripts that they do not truly match (the read is longer/shorter than the reference transcript, the read has different splicing in the middle compared to the reference transcript). It is most likely that you are getting a lot of false positives when directly aligning to the annotation.
FLAIR quantify essentially aligns your reads to a reference, then filters the alignments and only returns transcripts where the aligned reads truly match them. If you are suspicious that you are losing important genes/transcripts in flair collapse (also, we now recommend flair transcriptome instead of correct/collapse), you can try generating a bed + fasta file from your annotation gtf and then running flair quantify against the annotation.

[lgraciamaz]

Thank you very much for the response to my concerns. I will go ahead and try your suggestion.

Best,

Luis