🧬 GWAS Bioinformatics Tools

🧬 1. What is GWAS?

A Genome-Wide Association Study (GWAS) tests hundreds of thousands to millions of genetic variants (usually SNPs — Single Nucleotide Polymorphisms) across the genome to see whether any variant is statistically associated with a trait or disease.

Each SNP is tested independently, producing a p-value. Because so many tests are performed at once, raw p-values must be corrected for multiple testing — that's why GWAS uses thresholds like p < 5 × 10⁻⁸.

📊 2. SNP Statistics — HWE & Allele Frequency

Input format

One SNP per line: SNP_ID nAA nAa naa, where AA, Aa, aa are the three genotypes.

Hardy-Weinberg Equilibrium (HWE)

HWE states that in a large, randomly-mating population with no selection, mutation, or migration, genotype frequencies remain stable across generations. We test it with a χ² goodness-of-fit:

• Expected AA = p²·N
• Expected Aa = 2pq·N
• Expected aa = q²·N

where p = (2·nAA + nAa) / (2N) and q = 1 − p.

How to interpret HWE p-value

Allele frequency & MAF

p(A) = (2·nAA + nAa) / (2N). The Minor Allele Frequency (MAF) is min(p, q).

🧪 3. Case-Control Association

Input format

SNP_ID case_AA case_Aa case_aa ctrl_AA ctrl_Aa ctrl_aa

Genotypic χ² test (2 df)

Compares the 3 genotype counts between cases and controls using a 2×3 contingency table. A small p-value means genotype distribution differs between groups.

Allelic χ² test (1 df)

Collapses genotypes to allele counts (A vs a) and performs a 2×2 χ². More powerful when the effect is additive.

Odds Ratio (OR) & 95% CI

OR quantifies effect size. Using the dominant model (AA + Aa vs aa):

OR = (case_carriers · ctrl_non_carriers) / (case_non_carriers · ctrl_carriers)

Worked example

Suppose rs12345 gives OR = 2.10, 95% CI = 1.30 – 3.40, p = 0.002.

• OR > 1 → carriers of the risk allele have 2.1× higher odds of disease.
• CI (1.30–3.40) excludes 1.0 → significant association.
• p = 0.002 < 0.05 → statistically significant — but not genome-wide significant.

📐 4. Multiple Testing Correction

If you test 1,000,000 SNPs at α = 0.05, you expect ~50,000 false positives by chance. Corrections control this.

Bonferroni correction

p_adj = p × m (m = number of tests), capped at 1.0.

• Very strict — controls family-wise error rate (FWER).
• Overly conservative when tests are correlated (common in GWAS due to LD).

Benjamini-Hochberg FDR

Controls the expected proportion of false discoveries among significant results.

• Ranks p-values, then p_adj = p × m / rank, enforcing monotonicity.
• More powerful than Bonferroni — preferred in GWAS.

How to interpret the results table

📈 5. Manhattan Plot Data

Input format

SNP_ID Chromosome BP P_value — one row per SNP.

What is a Manhattan plot?

A scatter plot with genomic position on the X-axis (chromosomes laid end-to-end) and −log₁₀(p) on the Y-axis. Points that rise high above the threshold form "skyscrapers" — these are candidate associations.

Significance thresholds

Reading the output table

• −log₁₀(P) — the Y value you'd plot; higher = more significant.
• Status — automatically labelled based on thresholds.
• Rows are sorted by chromosome and position → ready for plotting in R, Python, or Excel.

✅ 6. Recommended Workflow

1. Quality control (QC) — filter SNPs with call rate < 95%, MAF < 0.01, or HWE p < 0.001 (in controls).
2. Tab 1: SNP statistics — verify HWE and MAF for each SNP.
3. Tab 2: Association testing — run χ² and OR for each SNP between cases and controls.
4. Tab 3 or auto-FDR in Tab 2 — apply FDR correction.
5. Tab 4: Manhattan data — format for plotting.
6. Follow-up — replication in an independent cohort; functional annotation.

⚠️ 7. Common Pitfalls & QC Filters

Reporting checklist

• Sample size (cases / controls)
• Number of SNPs tested
• QC filters applied
• Correction method used (FDR / Bonferroni)
• Effect size with 95% CI, not just p-value
• Replication status