Hello,
When pseudo-aligning bulk RNAseq fastq files with kallisto, I end up with very variable rates between my samples, sometimes below 30%.
When running kallisto on a set of fastqs, I get less than 10M reads pseudoaligned out of more than 40M reads Estimated average fragment length is 114.862, which is apparently way too short
Where can this come from ? Could it be adapter contamination ? I built the index using the transcript sequence fasta file from GENCODE
Is that much of a problem, if my downstream analysis looks normal, which genes where they're expected to be ? Though the problem is that the alignment is highly variable, I have samples over 50%, others below 20%...
Thank you in advance
1 answer
If so many reads aren’t aligning, you should pull out 10 or so reads at random, and run the sequences through BLAST.
Kallisto aligns to the mature mRNA transcriptome; your results suggest that you have reads not belonging to it. Your reads may be originating from introns, rRNA, some other species, etc.
Log in to answer this question.
-
STAR vs. Kallisto
written by gogeni5529 •
-
what is the purpose of indexing the reference genome (Kallisto)
written by Aaliya •
-
Troubleshooting RNA-seq data with DNA contamination
written by DKA •
-
Is there anything we can do when we have assignment rate below 50% in featurecounts?
written by Sib •
-
mapping reads using kallisto - rna seq analysis
written by bioinformatics •
-
Kallisto: Estimated average fragment length
written by jayeshkumarsundaram •
-
Kallisto quant error "std::bad_alloc"
written by bry.th •
-
kallisto genomebam won't write bam files
written by sstewar2 •
-
Using Gencode transcript annotation in Kallisto / Sleuth pipeline
written by Jordan •
-
salmon, kallisto psuedoaligners are failing to process
written by dr.prasad.bvls •
Hi,
Check your fastq files with a tool like fastqc. It displays if you have adapter contamination. It also show you if you have over-represented sequences which might indicate a different type of contamination. Check also, if your reads' quality is comparable between samples/runs.
EDIT: You also might want to check if the rRNA depletion/ mRNA-enrichment worked by aligning the reads against you target species' rRNA and mt-rRNA sequences.
FastQC appears to show a large percentage of transposase sequence and adapter towards the end of the reads, that's probably the source of the problem, is it mandatory to trim it ?
Trimming depends on your workflow. Standard-aligner like minimap2, STAR, etc. would soft-clip your reads' adapter sequences, rendering the trimming step obsolete. I don't know if that's the case for pseudo-aligner like Kallisto.
I would nevertheless trim the data if you are still in an exploratory phase, in order to get a feeling for your data. Make fastqc analyses for raw and trimmed data and check it with multiqc.