From Reads to a Count Matrix

The missing middle

The NGS track took you from raw reads to an aligned BAM file. The RNA-seq flagship starts from a finished table of gene counts. This lesson is the bridge between them: how sequencing reads become that table. It is the step almost every beginner is unsure about, and once you see it, the whole RNA-seq pipeline clicks together.

1What is a "count"?

A count is simply the number of sequencing reads that came from a given gene in a given sample. More reads from a gene means that gene was more highly expressed. To turn reads into counts, software needs two things: your reads, and an annotation (a file that says where each gene sits on the genome). It then assigns each read to the gene it overlaps and tallies them up.

2The two routes from reads to counts

There are two standard ways to get counts. They reach the same kind of table by different paths.

Route A: align, then count

This is the classic route. You align reads to the genome (Track 2 taught the idea), then hand the BAM and an annotation to featureCounts, which assigns each read to a gene.

# align (a splice-aware aligner; STAR shown conceptually), then count
featureCounts -a annotation.gtf -o counts.txt sample1.bam sample2.bam
# counts.txt now has one row per gene and one column per sample

Route B: quantify directly (the fast modern default)

salmon skips the slow step of placing every read on the genome. It matches reads straight against the list of known transcripts (the transcriptome), a trick called pseudo-alignment, then you collapse transcript estimates to gene-level counts in R with tximport.

# build an index once, then quantify each sample
salmon index -t transcripts.fa -i salmon_index
salmon quant -i salmon_index -l A -r sample1.fastq.gz -o sample1_quant
# then in R: tximport() reads the quant files into a gene-level count matrix

3Which route should a beginner use?

	Route A: align + featureCounts	Route B: salmon
Speed / compute	Slower, needs more memory	Fast, light, laptop-friendly
You also get	A genome alignment (BAM) you can view in IGV	Just the quantification
Good when	You need the alignment too, or study novel transcripts	You only need expression counts (most of the time)

For a first RNA-seq analysis, salmon is the friendliest default. If you have already aligned your reads for another reason, featureCounts on the BAM is just as valid.

4Why raw counts, not TPM, go into DESeq2

You will also see expression reported as TPM, CPM, or FPKM. These are normalized values, adjusted so you can compare a gene fairly across samples or against other genes. They are great for looking at and plotting, but differential-expression tools like DESeq2 want the raw integer counts, because the statistical model needs to know the actual number of reads (the count size carries the uncertainty). DESeq2 does its own normalization internally. So: raw counts in, and let the tool normalize.

Check your understanding

Now you have a count matrix, analyze it

Your First RNA-seq Analysis →