QC is not a beauty contest for plots. The real question is whether your data are trustworthy enough for the biological task ahead, and if not, which intervention is justified. Good QC combines plot reading, sample context, and a bias toward minimal but effective preprocessing.
fastqc and multiqc for standardized reportscutadapt, fastp, and trimmomatic for trimming and filtering# Generate QC reports
fastqc -t 4 sample_R1.fastq.gz sample_R2.fastq.gz
multiqc .
# Adapter trimming (illustrative)
cutadapt -j 4 -a ADAPTER_FWD -A ADAPTER_REV \
-q 20,20 --minimum-length 50 \
-o trimmed_R1.fastq.gz -p trimmed_R2.fastq.gz \
sample_R1.fastq.gz sample_R2.fastq.gz
Phred scores compress error probabilities into an additive scale:
Q = -10 * log10(p_error)That means Q30 is not a little better than Q20-it is ten times lower error probability. Still, trimming solely to maximize average Q can destroy useful data. Think in terms of downstream risk, not cosmetic improvement.
Trim adapters decisively because they create artificial sequence content.
Quality trimming should be conservative when read length is already precious.
Click through the modules below. The goal is to learn what each panel means, what common failure patterns look like, and when the correct answer is "investigate more" instead of trimming blindly.
Look for where median quality falls and whether the lower tail becomes dangerously broad near the end of reads.
Late-cycle increase usually means inserts were shorter than the read length, so sequencing reads into adapter sequence.
A single clean peak is not always required, but unexpected extra peaks often justify contamination checks or metadata review.
High duplication can reflect targeted sequencing, over-amplification, or low-complexity libraries. Interpretation depends on context.
These may be adapters, primers, contamination, or genuinely abundant biological reads. Sequence identity matters.
A gentle end-of-read decline is normal. What matters is whether the low tail becomes severe enough to disrupt mapping, overlap merging, or variant confidence.
Adapter sequence introduces false bases. Removing it is usually high value because it directly improves alignment and feature assignment.
Amplicon, capture, metagenomic, and RNA-seq libraries naturally have different GC behavior. The question is whether the observed profile matches the expected biology and protocol.
Whole-genome sequencing, RNA-seq, targeted sequencing, and amplicon data have different expectations. Use duplication together with library complexity and biological context.
Always identify what the sequences are before deciding what to do. Some are technical artifacts; others may be real biology such as rRNA or highly abundant transcripts.
Median, upper, and lower quantiles reveal when only a subset of reads is deteriorating versus when the whole run degrades.
A late-cycle surge is typical when fragments are shorter than the read length and sequencing enters adapter sequence.
This view helps separate healthy complexity from libraries dominated by repeated molecules.
| Observation | Likely cause | Good next step |
|---|---|---|
| Quality collapses only at the tail | Typical end-of-read decay | Consider light trimming or leave untouched if downstream tools tolerate it |
| Strong adapter rise late in reads | Short inserts | Trim adapters and confirm post-trim length distribution |
| Unexpected GC peak | Contamination or mixed library composition | Screen contamination and revisit sample metadata |
| Very high duplication in unbiased library | Low complexity / over-amplification | Assess whether re-sequencing or better library prep is needed |
| Same overrepresented sequence across many samples | Technical contaminant | Map the sequence identity before trimming or discarding |
If quality is poor from the beginning or oscillates in unusual patterns, no amount of trimming will create trustworthy data.
If most reads are duplicates, your effective unique information is far smaller than the total read count suggests.
If contaminant signatures dominate large fractions of the library, downstream interpretation may become more misleading than useful.