How much adapters is it ok to keep in my samples after RNA seq?

Question

[bioinfo]

Hello,

The software I am using for demultiplexing allows to specify adapters in the samplesheet so that the adapters can be trimmed. However, I noticed that when I did multiqc on my fastq files there are adapters left. I am including some examples below. In this example the adapter content is pretty low but I have had other samples with 3-7% adapters on the multiqc file.

When I specify the adapter exactly as Illumina says on their manual the adapter can reach up to 0.74%. The software also allows for automatic detection of the adapters. When I use that the adapter sequence that is picked up is the illumina adapter sequence+ 5 more bases. When I use that for adapter trimming multiqc says: No samples found with any adapter contamination > 0.1%

I wanted to know what would be the more appropriate way to trim the adapters? Would the extra 5 bases cause an issue?

Thank you

Answer 1

[GenoMax]

I wanted to know what would be the more appropriate way to trim the adapters? Would the extra 5 bases cause an issue?

This may depend on the end use application for the data. If you are planning to align to a good reference then most aligners will take care of any residual bases from adapter, that will not align, by soft-clipping.

For de novo work all extraneous sequence needs to be removed so you will need to use a separate scan/trim program such as bbduk.sh/fastp. A dedicated program is the best option to do this.

Running external scanning/trimming is a "one time and done" operation. That will leave you with guaranteed clean data that can be used for any application.

[bioinfo]

Thank you. I am planning to align the data with kallisto. Does that make a difference?

[GenoMax]

No it should not make a difference.

[bioinfo]

Thank you. I jut want to confirm. By not making a difference do you mean that trimming the extra 5 bases should not make a difference?

[GenoMax]

You could find out by running a sample with and without trimming. Passing the data through a trimming program is not going to add a great computational burden. Trim the data once and be sure that there is no extraneous sequence.

[bioinfo]

I had compared trimming with the illumina sequence which did not seem to work vs trimming with the illumina sequence+the 5 bases which seems to have trimmed everything based on fastqc. I had done a scatterplot with a pearson correlation for the gene counts and the tpm counts and the pearson correlation was 1. Do you think this is an appropriate way to compare them?

[GenoMax]

See: Quality filtering prior to pseudoalignment

This addresses trimming as well.