Genomics: Read Alignment, Variant Calling, and GWAS

BWA · GATK HaplotypeCaller · GWAS QC · Association testing · Manhattan plots

Published 2026 · Last updated 2026-09-27 · Hailu Zhou

The short-read genomics pipeline — align, call, filter, test — is over a decade old, yet most failures still trace to a skipped quality-control step. This post walks the canonical path from FASTQ to a defensible GWAS, with emphasis on the decisions that actually change results.

Alignment with BWA

BWA remains the reference short-read aligner. BWA-MEM handles reads from roughly 70 bp upward, performs local alignment, and emits SAM records with mapping quality that downstream tools depend on. BWA-MEM is less robust than a variation graph for highly divergent or structural regions, so keep expectations calibrated: reads crossing large indels or in segmental duplications are frequently mis-mapped, and those are precisely the regions where false-positive variants cluster.

Variant calling with GATK HaplotypeCaller

GATK's HaplotypeCaller performs local de novo assembly and realigns reads against candidate haplotypes, which sidesteps the reference-bias problems of naive pileup callers. In a cohort, use GVCF mode per sample and joint genotype calling, so genotype likelihoods are shared across samples and low-coverage sites are called consistently. Filter with variant quality score recalibration (VQSR) when you have enough known sites, or hard filters for small sets. Separate SNPs and indels, and always retain the unfiltered calls for reanalysis.

Why GWAS quality control dominates everything

A GWAS is mostly QC. Association statistics amplify every artifact; a single mislabelled batch or a handful of related individuals can manufacture genome-wide hits. The standard preflight:

Step Typical filter
Sample missingness< 2–5%
Variant missingness / MAF< 2%, MAF > 1%
HWE deviationp < 1e-6 in controls
RelatednessPI_HAT > 0.185 removed
Ancestry outliersPCA distance

Population structure is not optional. Principal components or a mixed model (GEMMA, BOLT-LMM, SAIGE) control both ancestry and, in the mixed-model case, cryptic relatedness. Report the inflation factor and lambda; a lambda near 1.0 after correction is the minimum credibility bar.

Association testing and effect-size reality

For quantitative traits, linear regression with covariates; for binary traits, logistic regression — but beware case-control imbalance, which inflates false positives in standard logistic models and motivates saddlepoint approximations such as SAIGE. The genome-wide significance threshold is conventionally p < 5e-8, derived from an estimate of roughly one million independent common variants in European-ancestry samples. That threshold is a family-wise convention, not a law of nature; trans-ancestry and rare-variant studies need their own calibration.

Reading a Manhattan plot honestly

A Manhattan plot compresses per-variant p-values on a genomic axis, with the y-axis as −log10(p). Use it to see locus-level clustering, not to rank individual variants. Always pair it with a quantile-quantile plot, because the Manhattan plot cannot tell you whether your inflation is global or concentrated in a few true loci. Fine-mapping and credible sets, not the lead SNP, are the correct output of a hit.

From hit to mechanism

A genome-wide hit marks a region; it rarely names the causal gene. Link the lead variant to functional annotation, expression or protein QTLs, and chromatin state. This is exactly where genomics hands off to colocalisation and Mendelian randomisation, and where most published claims overreach. Keep the handoff disciplined: association is not causation, and proximity to a gene is not evidence about it.

References

  1. Li H, Durbin R. Fast and accurate short read alignment with Burrows–Wheeler transform. Bioinformatics. 2009;25(14):1754–1760.
  2. Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv. 2013;1303.3997.
  3. McKenna A, Hanna M, Banks E, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20(9):1297–1303.
  4. DePristo MA, Banks E, Poplin R, et al. A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat Genet. 2011;43(5):491–498.
  5. Purcell S, Neale B, Todd-Brown K, et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet. 2007;81(3):559–575.
  6. Price AL, Patterson NJ, Plenge RM, et al. Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet. 2006;38(8):904–909.
  7. Zhou W, Nielsen JB, Fritsche LG, et al. Efficiently controlling for case-control imbalance and sample relatedness in large-scale genetic association studies. Nat Genet. 2018;50(9):1335–1341.
  8. Visscher PM, Wray NR, Zhang Q, et al. 10 years of GWAS discovery: biology, function, and translation. Am J Hum Genet. 2017;101(1):5–22.
  9. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76–82.