Discrepancy between number of reads from collapse --generate_map and quantification module output

[pgupta3005]

Hi, I used FLAIR correct, collapse and quantification modules after aligning Nanopore fastq files to the respective genome using minimap2. I used the following commands -

# FLAIR collapse
python /home/pallavi/softwares/flair-1.5/flair.py collapse \
-r PF/${i}.fastq \
-q $output_dir/${i}_all_corrected.bed \
-g $genome_fasta -f $genome_gtf \
--generate_map \
-t 12 --temp_dir /nfs_master/pallavi/tmp_work/ -o $output_dir/${i}_collapse 2> $output_dir/${i}_collapse.log

# FLAIR quantify
python /home/pallavi/softwares/flair-1.5/flair.py quantify \
-i $output_dir/${i}.isoforms.fa \
-r manifest_${i} \
-t 16 --tpm \
--temp_dir /nfs_master/pallavi/tmp_work -o $output_dir/${i}_quantify 2> $output_dir/${i}_quantify.log

While for most of the identified isoforms, the number of reads mapping to that isoform and the counts in the quantification output match, but for some the quantification returned is higher. For example -

As you can see, the number of reads mapped here are 5, but quantification gives 6. Why could this be happening and what is the solution?

Thanks!
-Pallavi

[pgupta3005]

Just to add - from a total for 2815 isoforms identified for our sample - the numbers matched for 1955 cases, while quantification gave higher counts for 787 isoforms and lower counts in 73 isoforms.

In IGV, it looks like this where the red ticks indicate the reads that were represented in the read-to-isoform map

[cafelton]

I think this is not an issue, just something that needs better documentation if I understand your situation correctly. If the mapped reads are in the isoform.read.map.txt file generated by FLAIR collapse and you're comparing them to FLAIR quantify, is is possible for these reads to not match up. This is because the process of de novo isoform identification in FLAIR collapse is distinct from the isoform quantification in FLAIR quantify. They share a lot of similarities and should produce similar, but not necessarily exactly the same, results. If you have to pick one to trust, I would pick the quantify file. You can also tell quantify to produce a read.map.txt file if you need that for your downstream pipeline.