Skip to content
Tech Interview Prep
Technical interview guide

Statistical Genomics

The statistical machinery behind genome-wide association studies: testing millions of variants for association with a trait without drowning in false positives.

Read
50 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: PLINK 2 alpha 7.x; GWAS Catalog and summary-statistics standards; EIGENSTRAT, BOLT-LMM, LD Score, KING, imputation, TOPMed and PRS source publications reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-11A team runs an additive linear GWAS on log10-transformed fasting glucose (mg/dL) in 40,000 adults, with genotypes coded 0/1/2 copies of the effect allele. For rs1234 the fitted coefficient is beta = 0.018 (SE 0.004, p = 6e-6). A reviewer asks what this number licenses them to say. Which statement is correct?
MCQ-14A GWAS reports a lead SNP (p = 3×10⁻¹²) that is directly genotyped, common (MAF 0.3), and intronic. The team treats it as the causal variant. Why may it still not be causal?
MCQ-15A reference panel's haplotypes across markers M1 and M2 are h1 = A-C, h2 = A-C, h3 = G-T, h4 = G-T. A sample is genotyped A/G at M1 and is untyped at M2. Using this panel, what imputed dosage of the T allele at M2 does the sample receive?
MCQ-1A team runs a GWAS on 12,000 cases and 40,000 controls, all of European ancestry, using a logistic regression with the first 10 principal components as covariates. A reviewer points out that the phenotype was measured at a single clinic visit, while the genetic effect they want to report is the lifetime risk of the disease. Which change to the analysis is required before the regression can be interpreted as estimating that effect?
MCQ-12A team runs a GWAS of 800,000 common variants in 5,000 cases and 5,000 controls, testing each variant with a two-sided allelic test at α = 0.05. The scan returns 40,000 variants with p < 0.05; the smallest p-value is 2×10⁻⁶. What does the 40,000 count mean, and what should the team do next?
MCQ-5Two GWAS summary-statistic files for rs1234 are meta-analysed by a script that matches on variant ID and averages beta. File A: effect allele T, other allele C, beta = +0.08, SE = 0.02, EAF = 0.30. File B: effect allele C, other allele T, beta = +0.08, SE = 0.02, EAF = 0.70. The script reports a pooled beta of +0.08. What does that pooled value mean?
MCQ-24A meta-analysis pipeline joins two GWAS summary-statistic files on rsID and combines betas without checking effect alleles. Both files are case-control studies of the same trait, both on GRCh37. At rs2000 (A/G), Study A reports effect allele A, beta = +0.09 (SE 0.02); Study B reports effect allele G, beta = +0.09 (SE 0.02). The pipeline reports a combined beta of +0.09 with p = 2×10⁻¹⁰. What is the correct interpretation?
MCQ-17Two European-ancestry cohorts (12,000 and 9,500 samples), genotyped on the same array and imputed to GRCh37, report a strong association at rs7412 with LDL cholesterol in mg/dL. Cohort A: effect allele T, beta = 10.2, SE = 0.9, p = 1e-29. Cohort B: effect allele C, beta = -9.8, SE = 0.9, p = 1e-27. A naive inverse-variance meta-analysis of the two betas returns beta = 0.2, SE = 0.64, p = 0.75. What is the correct next step?
MCQ-2A cohort of 2,400 samples was genotyped in two batches. After joint calling, QC reports that 37 samples have a heterozygosity rate more than 4 SD below the cohort mean, and those same 37 samples show a mean genotype concordance of 0.61 with their own array fingerprint (cohort median 0.998). The recorded sex for 12 of the 37 disagrees with the genotype-inferred sex. What is the most likely explanation for this pattern?
MCQ-4A GWAS QC pipeline runs a Hardy-Weinberg exact test on 12,000 unrelated controls of mixed self-reported ancestry. One common variant (MAF 0.31) returns p = 3e-9. Genotype counts are 5,412 / 5,180 / 1,408 (hom-ref / het / hom-alt), and the call-rate for this variant is 99.7% with no batch or plate effect. The same variant is in HWE in each ancestry stratum tested separately (all p > 0.05). What does the disequilibrium in the pooled sample most likely indicate?
MCQ-3A GWAS of 12,000 cases and 12,000 controls is genotyped on two plates. Plate 1 holds 9,000 cases and 3,000 controls; plate 2 holds 3,000 cases and 9,000 controls. Variant rs1234 has a call rate of 99.9% on plate 1 and 97.5% on plate 2, giving an overall call rate of 98.7% (cases 99.3%, controls 98.1%). It clears the study's 98% overall call-rate filter and shows OR = 1.38, p = 2×10⁻⁸. What does the case-control call-rate gap most likely indicate, and what should be done?
MCQ-10A GWAS of a binary disease fits, per variant, logistic regression of case status on genotype (0, 1 or 2 minor alleles), age, sex and 10 principal components, using 1,200 cases and 1,200 controls. At one rare missense variant the genotype counts are 2 / 13 / 1,185 among cases and 0 / 0 / 1,200 among controls, and the run reports convergence after 25 iterations with a genotype coefficient of 22.7, a standard error of 186.2 and a two-sided Wald p-value of 0.90. Which conclusion about this variant is correct?
MCQ-16A GWAS imputes 10,000 samples against 1000 Genomes phase 3. The team applies a global INFO ≥ 0.7 filter. A variant with MAF 0.8% has INFO 0.74 and passes. Its association p-value is 5×10⁻⁸, but it fails replication in an independent 8,000-sample cohort. Which QC change best addresses this result?
MCQ-7A team runs PCA on 8,000 samples using 700,000 array variants. PC1 explains 9% of the variance, and its largest loadings sit in a 4 Mb block on chromosome 8 that contains a common inversion; PC1 correlates with inversion genotype (r = 0.85) but barely with self-reported ancestry (r = 0.12). PC2 correlates with ancestry (r = 0.88). They plan to use PC1–PC10 as GWAS covariates. What does the pattern indicate?
MCQ-6A type 2 diabetes case-control study recruited 1,200 cases and 1,200 controls in one city: cases through a diabetes clinic, controls through a community screening programme. Ancestry analysis puts the cases at 30% West African and 70% European ancestry, and the controls at 60% West African and 40% European. At rs12345 the G allele frequency is 0.80 in the West African ancestry group and 0.30 in the European ancestry group, and inside each group its frequency is the same in cases and controls - the marker has no effect on diabetes in either stratum. The primary analysis pools all 2,400 people and adjusts for nothing. What per-allele odds ratio for allele G will that unadjusted analysis report?
MCQ-8A GWAS of adult height genotypes 20,000 people and reports rs12345 at β = 0.42 cm per allele, SE = 0.05, p ≈ 1e-16. Of those 20,000, 3,000 belong to 500 large sibships in a founder population (six genotyped siblings each on average); the other 17,000 are unrelated. Refitting the same SNP with a mixed model that includes a genetic relatedness matrix gives β = 0.41 and SE = 0.20. Which conclusion does the change in SE support?
MCQ-9A GWAS of 5,000 individuals fits a linear mixed model whose genetic relationship matrix is built from all 600,000 genotyped SNPs. Genomic control comes out at λGC = 0.82, and a chromosome 6 locus that reached p = 2×10⁻¹² in an unrelated replication cohort only reaches p = 4×10⁻⁵ in this analysis. What is the most likely cause, and what should the team do?
MCQ-20Two GWAS of the same disease, each with 150,000 cases and 150,000 controls, are analyzed with univariate LD score regression using the same ancestry-matched LD scores. Study A returns an intercept of 1.38 (SE 0.03); Study B returns 1.02 (SE 0.02). Both slopes convert to a similar liability-scale SNP heritability. What does the difference in intercepts tell the team?
MCQ-13A GWAS of height in 50,000 adults from one ancestry excludes related pairs (kinship > 0.088) and adjusts for 10 principal components. The genomic inflation factor is lambda GC = 1.32. LD Score regression gives an intercept of 1.01 (SE 0.008). The QQ plot departs from the null only for P < 1e-5. What should the analyst conclude?
MCQ-18A meta-analysis pools 12 cohort studies of a variant–biomarker association. Fixed-effect pooled OR = 1.18 (95% CI 1.10–1.27); Cochran's Q = 47.3 on 11 df (p < 0.001); I² = 77%; τ² = 0.04 on the log-OR scale. Random-effects pooled OR = 1.09 (95% CI 0.94–1.26). What does this pattern indicate?
MCQ-19Twenty thousand exomes, 10,000 cases and 10,000 controls, sequenced across a gene G. No single variant in G has more than 40 carriers. Functional work classifies all 13 protein-truncating (PTV) variants and all 22 rare activating missense variants in G as risk-increasing, though the missense effects are far weaker. Three collapse-based tests use masks built from annotations fixed before the analysis. Mask 1, the 13 PTVs, burden test: 1.0% of cases (100/10,000) and 0.5% of controls (50/10,000) carry at least one masked variant, OR = 2.0, p = 4e-5. Mask 2, the 13 PTVs plus the 22 missense variants, burden test: 4.0% of cases (400/10,000) and 3.5% of controls (350/10,000), p = 0.06. Mask 3, the same 41 variants, SKAT-O: p = 0.02. Which account of the three results is right?
MCQ-21A team uses rs123 as an instrument for LDL cholesterol in a Mendelian randomization study of myocardial infarction. In 120,000 samples, rs123 is strongly associated with LDL (F = 48) and with MI (OR 1.12 per allele). A separate GWAS in the same population finds rs123 also associated with smoking initiation (OR 1.06), and smoking independently raises MI risk. What does this imply for the MR estimate?
MCQ-23A PRS for type 2 diabetes includes rs111111 with a weight (log OR) of 0.30, estimated in a European-ancestry GWAS. In a target African-ancestry cohort, rs111111 has an allele frequency of 0.22 (vs 0.30 in Europeans) and is in weak LD (r² = 0.08) with the causal variant at that locus, which is tagged by rs222222 in the target population. The PRS does not include rs222222. What is the expected contribution of rs111111 to the PRS in the target cohort?
MCQ-22A team ships a polygenic score for coronary artery disease. Weights come from a 450,000-participant European-ancestry GWAS meta-analysis. On 8,000 held-out European-ancestry samples the score reaches AUC 0.71 with calibration slope 0.98; on 3,000 South Asian patients at a partner clinic it reaches AUC 0.58 with calibration slope 0.61, and the top-decile cutoff flags 34% of that cohort against 10% in validation. The clinic needs absolute 10-year risk numbers for its patients, and a South Asian GWAS (n = 60,000) is available. Which action is most defensible?
MCQ-25A PRS for coronary artery disease, trained in a European-ancestry cohort, is being validated in a multi-ancestry cohort whose baseline 10-year incidence and linkage-disequilibrium structure differ from the training cohort's. The team must report two things separately: how much the PRS adds to a clinical model already containing age, sex, and ancestry principal components, and whether the score's predicted absolute risks transfer to this cohort's baseline risk. Which assessment strategy answers both questions?