Library prep and sequencing: anatomy of a read
The amplified cDNA is full length, far too long for a short-read sequencer. Library preparation cuts it, adds sequencing adapters and a sample index, and makes sure that only the barcoded 3′ end of each transcript becomes a sequenceable molecule. The Illumina sequencer then reads each fragment from both ends: Read 1 gives the barcode and UMI, Read 2 the cDNA itself.
From cDNA to a sequencing library
Library preparation has roughly four steps:
- cDNA amplification and QC. After PCR, the size distribution and yield are checked. Low cDNA yield often means few cells or RNA-poor cells.
- Fragmentation. An enzyme cuts the long cDNA into pieces of a sequenceable length.
- End repair, A-tailing and adapter ligation. An adapter carrying the Read 2 primer sequence is ligated to the fragment ends.
- Sample index PCR. This adds the P5/P7 sequences that bind the flow cell, and the sample index that tells libraries apart.
Step 4 is the key one. Only fragments with the bead primer (Read 1 sequence, barcode, UMI) at one end and the Read 2 adapter at the other are amplified correctly by this PCR. Fragments from the middle of a cDNA have no Read 1 sequence and do not amplify. The final library is therefore almost entirely transcript 3′ ends, hence the name “3′ sequencing”.
Sequencing by synthesis: anatomy of a read
On an Illumina flow cell, both ends of each library molecule bind primers on the surface, and bridge amplification grows a cluster of thousands of identical copies. In each sequencing cycle, one fluorescent nucleotide with a temporary block at its end is added; a camera records each cluster’s colour, the dye and block are removed, and the next cycle begins. One cycle reads one base.
With 3′ v3/v3.1, each library molecule is read in several segments:
- Read 1 (28 cycles): 16 bp cell barcode + 12 bp UMI. It contains no gene information.
- i7 / i5 index: the sample index, 8 bp for single-index kits, 10 bp each for dual-index kits.
- Read 2: the cDNA sequence used to map to genes, commonly about 90 bp (follow the kit’s guide).
The v1 libraries behind pbmc3k were laid out differently: according to 10x’s dataset description, the cDNA was in a 98 bp Read 1, the 14 bp barcode came from an index read and the 10 bp UMI from Read 2. When comparing data from different years, do not assume the v3 layout.
Sequencing quality is measured as Q30: base quality ≥ 30, an error rate of at most 0.1%. Quality usually falls as the read gets longer.
Common pitfalls
- Read 1 too short. Read 1 must cover the full barcode and UMI; a few cycles short and the UMI is truncated.
- Looking only at total reads. The numerator of Mean Reads per Cell includes empty-droplet reads; with a lot of ambient RNA, cells actually receive less depth.
- Assuming deeper is always better. Once saturation is high, extra sequencing finds almost no new molecules.
- Ignoring index hopping when pooling. When many samples share a patterned flow cell, use unique dual indexes.