From Reads to Variants: An Introduction to NGS Data Analysis Pipelines

Whether it's WES, WGS, or a targeted panel, most clinical NGS pipelines follow the same broad shape. Here's the map before the detail.

1. Raw reads (FASTQ)

The sequencer outputs millions of short reads, each with a quality score per base. See the file formats note for the format itself.

2. Quality control and trimming

Before doing anything else, check the data. FastQC reports per-base quality, adapter contamination, and GC content at a glance. Low-quality bases and leftover sequencing adapters are then trimmed with tools like fastp or Trimmomatic - feeding bad data into an aligner just produces bad alignments.

3. Alignment

Reads are mapped to a reference genome (e.g. GRCh38) with an aligner - BWA-MEM is the long-standing standard for short reads. The output is a BAM file: every read's position, and how well it matches.

4. Variant calling

A caller compares the aligned reads to the reference at every position and proposes differences. GATK HaplotypeCaller and DeepVariant (a deep-learning-based caller from Google) are two widely used options for germline variants; somatic (cancer) calling uses different tools entirely, since it has to distinguish real mutations from sequencing noise at low allele fractions.

5. Annotation

A raw VCF just says where the genome differs - it says nothing about what that means. Annotation tools (VEP, ANNOVAR, SnpEff) add gene names, predicted protein consequence, and population frequencies from databases like gnomAD.

6. Interpretation and reporting

Finally, a human (or a well-built review system) filters the annotated list down from thousands of variants to the handful that plausibly explain the patient's phenotype, classifies them under ACMG/AMP guidelines, and writes the clinical report.

Further reading

• GATK Best Practices - the reference workflows for germline and somatic calling
• nf-core - a large library of peer-reviewed, community-curated NGS pipelines
• Ensembl VEP - variant annotation