Bio Ribo Seq Riboseq Preprocessing

Version Compatibility

Reference examples tested with: cutadapt 4.4+, umi_tools 1.1+, STAR 2.7.11+, bowtie2 2.5.3+, SortMeRNA 4.3+, samtools 1.19+

Before using code patterns, verify installed versions match. If versions differ:

• CLI: <tool> --version then <tool> --help to confirm flags
• Python: pip show <package> then help(module.function) to check signatures

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Ribo-seq Preprocessing

"Preprocess my ribosome profiling data" -> Extract UMIs, trim the 3' linker, deplete rRNA/tRNA contaminants, align footprints with end-to-end (non-soft-clipped) settings, deduplicate only when UMIs allow it, and QC the read-length distribution.

• CLI: umi_tools extract -> cutadapt -> bowtie2/SortMeRNA (contaminant removal) -> STAR (genome + transcriptome projection) -> umi_tools dedup -> samtools

The canonical modern order (nf-core/riboseq, McGlincy & Ingolia 2017) is UMI-extract FIRST (the UMI lives in the read and must move to the read name before the linker is cut), then trim, then contaminant removal (before the expensive aligner), then align, then dedup on the BAM.

Upstream context that changes the analysis (ask before trusting the data)

• How were cells harvested, and with which drug? Cycloheximide (CHX) pre-treatment of live cells lets initiation continue while elongation arrests, fabricating start-codon and 5'-ramp density and distorting downstream dwell-time work (Hussmann 2015). Flash-freeze with no drug (or CHX only in the lysis buffer) is the gold standard. Harvest method is recorded at preprocessing because it gates which downstream conclusions are valid (see ribosome-stalling).
• Which nuclease? RNase I (eukaryotes) trims close to the ribosome with little sequence bias, giving sharp ~28-30 nt footprints and crisp periodicity. RNase I is inhibited by the E. coli ribosome and FAILS in bacteria, so bacterial protocols use micrococcal nuclease (MNase), which has sequence bias, broader footprints, and forces 3'-end P-site anchoring (Mohammad 2019). A eukaryote-tuned pipeline silently misanalyzes MNase/bacterial data.
• Are there UMIs? The dedup decision depends entirely on this (table below).

The decisions that shape preprocessing

Deduplication: with-UMI vs without-UMI (the load-bearing choice)

Alignment: genome (STAR, spliced) vs transcriptome (bowtie2, unspliced)

Contaminant removal approach

rRNA is the dominant contaminant: commonly 50-90% (often >80%) of a Ribo-seq library, because nuclease digestion of the ribosome itself produces abundant rRNA fragments in the footprint size range. Wet-lab depletion (RiboZero/RiboCop/biotinylated subtraction oligos) reduces but never eliminates it, so in-silico removal is mandatory. Effective mRNA depth is a small fraction of raw reads.

Extract UMIs

Goal: Move the UMI from the read sequence into the read name so it survives every later step and can deduplicate the final BAM.

Approach: Run umi_tools extract FIRST, before adapter trimming, with the barcode pattern matching the library's read structure (N = random UMI base extracted to the name, X = fixed base kept).

# Only when the library has UMIs. Pattern is library-specific.
# McGlincy & Ingolia 2017 split the 7-nt UMI (5 nt in the linker + 2 nt from circularization)
umi_tools extract \
    --bc-pattern=NNNNN \
    --stdin reads.fastq.gz \
    --stdout reads.umi.fastq.gz \
    --log umi_extract.log

When the UMI is split across the read (an inline 5' portion plus a portion inside the 3' linker, as in McGlincy & Ingolia 2017), the linker-embedded part is otherwise lost at trimming: extract it from the 3' end too (a second umi_tools extract with a --3prime pattern, or cutadapt's {N} linker capture) rather than discarding it. A pattern matching only the 5' inline bases recovers half the UMI and under-collapses duplicates.

Trim the 3' linker

Goal: Remove the 3' adapter that is always read through because footprints (~28-30 nt) are far shorter than the read.

Approach: Run cutadapt with the known adapter and a PERMISSIVE length floor, and discard reads where no adapter was found.

# --discard-untrimmed: a footprint without read-through adapter is almost never a real footprint
# -m 15: permissive floor (do NOT narrow to 28-32 yet; inspect the length distribution first)
cutadapt \
    -a CTGTAGGCACCATCAAT \
    --discard-untrimmed \
    -m 15 -M 40 \
    -j 0 \
    -o reads.trimmed.fastq.gz \
    reads.umi.fastq.gz

The classic Ingolia linker CTGTAGGCACCATCAAT is an example only; the real sequence is protocol/kit-specific and McGlincy-Ingolia linkers embed the UMI and sample barcode, so the trimmed "adapter" region may include them.

Remove rRNA and other contaminants

Goal: Discard rRNA/tRNA/snoRNA reads before the expensive spliced aligner runs.

Approach: Align to a combined contaminant index and keep only the unmapped reads, OR use a dedicated rRNA filter.

# Option A: combined contaminant index (rRNA + tRNA + snoRNA + snRNA), keep unmapped
bowtie2 -x contaminant_index \
    -U reads.trimmed.fastq.gz \
    --un-gz reads.noncontam.fastq.gz \
    -S /dev/null -p 8

# Option B: SortMeRNA v4 (use a per-sample --workdir; a shared kvdb collides across runs)
sortmerna \
    --ref rRNA_db/silva-euk-18s-id95.fasta \
    --ref rRNA_db/silva-euk-28s-id98.fasta \
    --reads reads.trimmed.fastq.gz \
    --aligned rRNA_hits --other reads.noncontam \
    --fastx --workdir sortmerna_sampleA --threads 8