Issues with adapter trimming (Trim Galore)

Question

I am new to coding and especially new to bioinformatics so I am sorry if this is a dumb question. Nevertheless, I am attempting to run Trim Galore! to trim my paired RNAseq data, there are no error messages that get thrown when trim is running. However, when I check the fastQC reports there are a ton of indicators showing the data is low quality. I am unsure of what I am missing and would really appreciate some help. The only thing I find odd is there are low adapter counts and I am not really sure why this is.

Found perfect matches for the following adapter sequences:

Adapter type	Count	Sequence	Sequences analysed	Percentage
Nextera	11	CTGTCTCTTATA	1000000	0.00
smallRNA	10	TGGAATTCTCGG	1000000	0.00
Illumina	0	AGATCGGAAGAGC	1000000	0.00

Using Nextera adapter for trimming (count: 11). Second best hit was smallRNA (count: 10)

This is the code I am running to run Trim Galore:

trim_galore --paired --fastqc --length 20 --adapter  -o /path/to/R1_file /path/to/R2_file

My FastQC reports fail at the following parameters:

• Per base sequence quality
• Per base sequence content
• Per sequence GC content
• Sequence Length Distribution

I also have a warning error at "Over represented sequences" with 17 sequences all with no possible sources. As I mentioned above any help or advice would be greatly appreciated.

UPDATE I ran FASTQC on my untrimmed fastq data and received the same errors/warnings. I have attached the plots below.

[3

Answer 1

1. Ignore the overrepresented sequences. The counts for these that you show is so low (10, 11) that it is irrelevant.

2. Ignore per-base sequence content. This in RNA-seq is known to be a bit messed up since the typical cDNA primer binding introduces some bias at the 5' end. I cannot really explain what the direct implication of GC content is, but can say with confidence that I really never really cared about this warning (or error) and never got into any trouble with it. Sequence length distribution gets warning if sequences are of different length, which is normal after trimming.

"Per base sequence content" is fine as well, it's only the first base that is a bit odd (not sure why) but I would not worry. Aligners will soft-clip it off if it causes mismatch alignments.

The question you need to ask is whether trimming is even necessary. My personal take on RNA-seq is that trimming only makes sense if you have adapter contamination. If not, skip it.

Answer 2

Looking at the plots, the only one I'd look into further is the per-sequence GC content, because the distribution is quite skewed in comparison to the theoretical distribution. This could be explained away if you're using a patterned flow cell with single-dye chemistry (e.g. NovaSeq), because their "no signal" sequence is a string of Gs (which would show up as a spike on a graph). Another explanation would be contamination with bacterial or other non-sample RNA; in many cases such contamination can be checked for by looking for ribosomal RNA in the reads.

Given that you're looking at cDNA reads, the per-base sequence content looks fine. cDNA reads often have an abundance of As and Ts near the ends thanks to the polyA tail, and the other slight over-representation near the end is G, which (as mentioned) might already be expected in Illumina reads.

As others have discussed, the overrepresented sequences are a low enough fraction that it's probably not going to be an issue with analysis.

I see that you've also included "Sequence Duplication Levels", that includes a statement "Percent of seqs remaining if deduplicated 40.79%". That level of duplication strikes me as odd, possibly pointing to something like a B-cell spitting out millions of copies of its receptor transcript - I use this as a demonstrative example because I've encountered it myself, not because I think that's happening with your data. Such a situation could also explain the GC content difference as well, because there will be less transcript diversity.