Advancedbiotechnology

Genome-Wide Association Studies

Overview

GWAS scans millions of SNPs across thousands of cases and controls, finding variants statistically associated with traits. Since 2005, GWAS has mapped 100,000+ associations: most in non-coding regulatory regions with small effects. Manhattan plots display significance; polygenic risk scores sum effects for prediction. Limitations: Eurocentric bias, missing heritability, correlation not causation.

Learning objectives

  • Explain the GWAS design
  • Read a Manhattan plot
  • Explain polygenic risk scores
  • Discuss limitations (ancestry bias, missing heritability)

The design

Genotype cases and controls at ~1M SNPs (imputed to ~10M+); test each SNP for frequency differences; genome-wide significance threshold p < 5x10^-8 (Bonferroni for 1M tests). Needs thousands of samples for small effects: modern GWAS use hundreds of thousands (UK Biobank).

5x10^-8

The genome-wide significance threshold: strict because a million tests are run.

Manhattan plots

Each SNP is a point: x = genomic position, y = -log10(p). Skyscrapers above the significance line are associated loci. Peaks usually span LD blocks; the causal variant is rarely the top SNP: fine-mapping follows.

Peaks, not pins

LD means associated regions, not single causal SNPs: the plot shows neighborhoods.

Polygenic risk scores

PRS sums risk alleles weighted by effect size: individuals in the top percentiles carry several-fold higher risk (coronary disease, breast cancer, diabetes). Clinical use is emerging but debated: PRS stratifies, rarely diagnoses, and performs worse in non-European ancestries.

Ancestry bias

~80% of GWAS participants are European: PRS accuracy drops sharply in other ancestries, a major equity issue.

Missing heritability and beyond

GWAS hits explain fractions of heritability: rare variants, structural variants, interactions, and overestimated twin heritability share the blame. Whole-genome sequencing and diverse biobanks (All of Us) are closing gaps; functional follow-up (eQTLs, CRISPR screens) turns loci into biology.

Association is not causation

A GWAS peak needs functional validation before it becomes a drug target.

Key points

  • GWAS tests millions of SNPs case-vs-control; p < 5x10^-8
  • Manhattan plots show associated loci (LD blocks, not causal SNPs)
  • PRS sums weighted risk alleles for stratification
  • Limits: Eurocentric data, missing heritability, non-causal associations

Common misconceptions

✗ Myth: GWAS finds the mutations that cause disease.

✓ Truth: It finds associated tag SNPs, mostly non-coding with small effects; causation needs functional proof.

✗ Myth: A high PRS means you will get the disease.

✓ Truth: PRS shifts probabilities modestly; most high-PRS people never develop the condition.

Practice questions

Q1. Genome-wide significance is conventionally
  1. p < 0.05
  2. p < 5x10^-8
  3. p < 0.5
  4. Any p
Answer: B.

Correcting for ~1M tests sets the stringent threshold.

Q2. Most GWAS hits are of which type?
  1. Protein-truncating
  2. Non-coding variants with small effects
  3. Chromosomal
  4. Mitochondrial
Answer: B.

Regulatory variants of modest effect dominate GWAS findings.

Q3. A major limitation of PRS is
  1. Too accurate
  2. Poor transfer across ancestries due to Eurocentric data
  3. Too cheap
  4. No math
Answer: B.

Scores trained in Europeans underperform elsewhere: an equity problem.

Click a question to reveal the answer and explanation.

Frequently asked questions

What is linkage disequilibrium's role?

LD lets genotyped tag SNPs represent ungenotyped neighbors, but blurs which variant is causal.

Why 5x10^-8?

Bonferroni correction for ~1M independent tests at alpha 0.05: controls false positives genome-wide.

What is Mendelian randomization?

Using GWAS SNPs as instrumental variables to infer causality (e.g., LDL causes heart disease): genetics as a natural trial.