Read processing & analysis
Quality control of raw reads with FastQC or similar tools: per-base quality, per-sequence quality, base composition, GC content, duplication, adapter content and over-represented sequences, ideally summarised across samples with MultiQC. It decides what trimming the data needs and flags failed runs.
Our ruleSimulations prepare you for the bench but don't replace hands-on practice.
Read QC (FastQC)
Quality control of raw reads with FastQC or similar tools: per-base quality, per-sequence quality, base composition, GC content, duplication, adapter content and over-represented sequences, ideally summarised across samples with MultiQC. It decides what trimming the data needs and flags failed runs.
How it differs from the standard method
- Some 'failures' are normal for a library type: RNA-seq shows high duplication and biased early cycles from random priming.
- Run it before and after trimming to see what changed.
- A contamination screen (FastQ Screen style) explains reads that map nowhere.
1From reads to results is a chain of decisions
A sequencer delivers millions of short reads with a quality score for every base, and nothing else. Turning them into a gene list, a set of variants or a peak file takes a pipeline: check the reads, remove what is not genomic sequence, place each read on a reference genome, filter the alignments, and then count or call.
Look before you align
FastQC-style reports show per-base quality, adapter content, base composition, duplication and over-represented sequences. They don't fix anything, but they tell you what trimming and filtering the data needs.
Place every read
DNA reads go to BWA-MEM or Bowtie2; RNA reads need a splice-aware aligner such as STAR or HISAT2 that can split a read across an intron. The reference build and the annotation must match.
Turn alignments into results
Variant callers weigh base and mapping qualities into genotype likelihoods; expression tools count reads per gene and model them as negative-binomial counts. Both depend on the filters before them.
2What trimming is for
When the insert is shorter than the read, the sequencer reads on into the adapter. Those bases are not in the genome.
3Calling variants honestly
A genotype caller assumes each read is independent evidence and that each base's quality is its true error rate. PCR duplicates break the first assumption, which is why they are marked.
4Testing expression
DESeq2 normalises samples with median-of-ratios size factors, which a handful of very strongly induced genes cannot distort the way total-count scaling can. It estimates each gene's biological dispersion, shrinks the estimates towards a trend across genes because a few replicates estimate variance poorly, fits a negative-binomial model with every known nuisance factor (such as batch) in the design, and tests with a Wald test followed by Benjamini–Hochberg adjustment.
Key terms
Practise these as flashcards →−10·log10 of the probability a base call is wrong: Q30 means 1 error in 1,000.
Sequencing past the end of a short insert into the adapter on the far side.
Leaving the ends of a read unaligned while aligning the rest.
Phred-scaled confidence that a read is placed correctly; 0 means it fits equally well elsewhere.
An RNA-seq read split across an intron, with each part on its own exon.
Base-quality score recalibration: re-estimating base qualities from mismatches at sites not expected to vary.
A per-sample scaling for sequencing depth and composition, used to compare counts between samples.
In a negative-binomial model, the squared biological coefficient of variation of a gene between replicates.
A Benjamini–Hochberg adjusted p-value; calling genes at padj < 0.05 controls the false-discovery rate at 5%.
When to use it
- Every sequencing dataset: QC, trimming and alignment come first whatever the question.
- Germline variant calling from whole-genome, exome or panel data.
- Differential expression between conditions from bulk RNA-seq counts with biological replicates.
Compared with alternatives
Where the reads and their quality scores come from; run metrics explain many QC findings.
The wet-lab side of an expression study: replicates and design are decided there, not in analysis.
Depth and coverage evenness set what variant calling can recover.
Insert sizes and duplicates seen in QC reports trace back to the library.
Sub-techniques
Choosing one here switches the whole page, including the protocol and the lab, to it.
Quality control of raw reads with FastQC or similar tools: per-base quality, per-sequence quality, base composition, GC content, duplication, adapter content and over-represented sequences, ideally summarised across samples with MultiQC. It decides what trimming the data needs and flags failed runs.
- Some 'failures' are normal for a library type: RNA-seq shows high duplication and biased early cycles from random priming.
- Run it before and after trimming to see what changed.
- A contamination screen (FastQ Screen style) explains reads that map nowhere.