Bio Read Qc Quality Filtering

Version Compatibility

Reference examples tested with: Trimmomatic 0.39+, fastp 0.23+, Cutadapt 4.4+

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

• CLI: <tool> --version then <tool> --help to confirm flags

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

Quality Filtering -- trim lightly or not at all, and never without a length filter

Trim low-quality bases and drop low-quality reads with Trimmomatic (sliding window / MAXINFO), fastp (per-read filter + window cut), or Cutadapt (BWA-style quality trim).

"Filter reads by quality" -> Remove low-quality bases and/or discard reads below quality/length thresholds.

• CLI: fastp -i in.fq -o out.fq --cut_right -q 20 -l 36 (window trim + per-read filter + length gate)
• CLI: trimmomatic SE in.fq out.fq SLIDINGWINDOW:4:20 MINLEN:36

Scope: this skill OWNS quality/length/N/complexity filtering. Adapter removal -> read-qc/adapter-trimming. Single-pass trim+QC -> read-qc/fastp-workflow. Reading the quality plots -> read-qc/quality-reports. OUT OF SCOPE: contamination removal (read-qc/contamination-screening).

The Single Most Important Modern Insight

1. Modern local aligners SOFT-CLIP low-quality tails, so quality trimming is usually unnecessary -- and AGGRESSIVE quality trimming actively harms downstream results. Williams 2016 showed aggressive trimming changed expression estimates for >10% of genes; Del Fabbro 2013 showed stringent Q>30 DEGRADES de novo assembly; MacManes 2014 found gentle trimming (remove only Phred<2-5) optimal for RNA-seq; GATK discourages quality trimming because BQSR recalibrates qualities itself. Trim ADAPTER always (read-qc/adapter-trimming); quality-trim lightly or not at all before a soft-clipping aligner. The workflows that genuinely need quality trimming are assembly, k-mer/pseudo-alignment, small-RNA, amplicon, and variant calling WITHOUT BQSR.

2. Quality FILTERING (drop whole reads) and quality TRIMMING (cut bases within a read) are different operations with different tools. fastp's -q/-u/-n filters whole reads by the fraction of unqualified bases; --cut_right / Trimmomatic SLIDINGWINDOW trims bases from a window scan. Filtering removes a junk subpopulation (a low-Q hump in the per-sequence-quality plot); trimming shortens reads with decayed tails. Choose by whether the problem is some bad reads or bad ends.

3. A short post-trim read mis-maps, so quality trimming MUST be paired with a minimum-length filter. Williams 2016 showed that adding a post-trim min-length filter mitigates most of the expression distortion that trimming introduces, because over-trimmed fragments that would map spuriously are dropped instead. MINLEN (Trimmomatic, always last), -l (fastp), -m (cutadapt) are not optional add-ons; they are the safety mechanism that makes trimming safe.

Two-color note: on NextSeq/NovaSeq the quality scores are binned to four values (RTA3: 2, 12, 23, 37), so a sliding-window threshold like 4:15 partitions between the 12 and 23 bins rather than acting on a smooth gradient -- thresholds tuned on HiSeq-era 0-40 qualities behave differently. And poly-G tails are HIGH quality, so a quality filter does not remove them (use poly-G trimming).

Tool Taxonomy

Decision Tree by Scenario

Default when uncertain: trim adapter, apply a light window trim plus a minimum-length filter, then confirm with FastQC.

Trimmomatic

Steps run in COMMAND-LINE ORDER; put quality steps before MINLEN so the length check reflects all trimming.

# Single-end: light leading/trailing + window, length-gated
trimmomatic SE -phred33 in.fq.gz out.fq.gz \
    LEADING:3 TRAILING:3 SLIDINGWINDOW:4:20 MINLEN:36

# Paired-end (four outputs: paired + orphan)
trimmomatic PE -phred33 -threads 8 \
    R1.fq.gz R2.fq.gz \
    R1_paired.fq.gz R1_unpaired.fq.gz R2_paired.fq.gz R2_unpaired.fq.gz \
    SLIDINGWINDOW:4:20 MINLEN:36

# MAXINFO: adaptive length-vs-quality balance (strictness <0.2 favors length, >0.8 favors correctness)
trimmomatic SE in.fq.gz out.fq.gz MAXINFO:40:0.5 MINLEN:36

Do NOT HEADCROP the first ~12 bp of RNA-seq to "fix" the wavy per-base-content plot: that pattern is random-hexamer priming bias (Hansen 2010), not adapter, and trimming it just discards real data without removing the underlying bias.