Explain genome-wide association studies (GWAS)

6.1.9 · Biology › Genomics

Date: 2026-07-01
Tags: #genomics #GWAS #complex-traits #statistical-genetics #SNP-analysis

Overview

Genome-Wide Association Studies (GWAS) ek systematic approach hai jisse genetic variants (specifically single nucleotide polymorphisms or SNPs) identify kiye jaate hain jo specific traits ya diseases se associated hote hain — poore genome mein cases aur controls ke beech allele frequencies compare karke.

[!intuition] Core Intuition

GWAS ko ek genetic treasure hunt samjho lakho jagahon par. Socho tum jaanna chahte ho ki kuch log tall kyun hote hain aur kuch short. Ek ya do genes check karne ki jagah, tum lakho se crore genetic positions ek saath check karte ho, kai logon mein.

Asli baat ye hai: Agar ek particular genetic variant (jaise chromosome 7 par position 12,345,678 par G ki jagah A hona) bahut zyada frequently tall logon mein milta hai short logon ke comparison mein, toh woh variant height se associated hai. Ye bilkul waise hai jaise notice karna ki red shirts wale log line ke tall waale end par cluster karte hain — red shirt (genetic variant) tallness se correlated hai, chahe woh directly cause na kare.

GWAS kyun important hai: Zyaadatar common diseases aur traits (diabetes, height, schizophrenia, blood pressure) polygenic hote hain — bahut saare genes se influence hote hain, har ek ka chhota effect hota hai. GWAS woh tool hai jo in subtle genetic contributions ko detect kar sakta hai.

[!definition] Formal Definition

Ek genome-wide association study (GWAS) alag-alag individuals mein genome-wide set of genetic variants ka ek observational study hai, jisse yeh determine kiya jaata hai ki koi variant kisi phenotype se associated hai ya nahin. GWAS typically single nucleotide polymorphisms (SNPs) aur major traits ya diseases ke beech associations par focus karta hai.

Key components:

Cases: Jin individuals ko disease/trait hai
Controls: Jin individuals ko disease/trait nahin hai
SNPs: Genetic markers jo test kiye jaate hain (typically 500,000 se 5+ million)
Association test: Allele frequencies ka statistical comparison

[!formula] Statistical Foundation

The 2×2 Contingency Table

Har SNP ke liye, hum allele frequencies compare karne ke liye ek table banate hain:

	Cases	Controls
Allele A	a	b
Allele G	c	d

Chi-Square Test of Association

$χ^{2} = \frac{n ( a d - b c ) ^{2}}{( a + b ) ( c + d ) ( a + c ) ( b + d )}$

jahaan $n = a + b + c + d$ (total allele count) hai.

Yeh formula kyun?

Numerator $(a d - b c)^{2}$ independence se deviation measure karta hai. Agar alleles cases aur controls mein identically distributed hain, toh $a d \approx b c$ , jisse $χ^{2} \approx 0$ ho jaata hai.
Denominator marginal totals se normalize karta hai, overall allele frequencies aur sample sizes ko account karte hue.
Null hypothesis ke under (koi association nahin), $χ^{2}$ ek chi-square distribution follow karta hai 1 degree of freedom ke saath.

P-value and Significance

$P = P (χ_{(1)}^{2} \geq χ_{o b ser v e d}^{2})$

Lekin hum lakho SNPs test karte hain, isliye humein Bonferroni correction chahiye:

$α_{cor r ec t e d} = \frac{α}{m}$

jahaan $m$ independent tests ki sankhya hai.

Genome-wide significance threshold ki derivation:

$α = 0.05$ se shuru karo (standard significance)
~1 million independent SNPs maano (linkage disequilibrium account karte hue)
$α_{G W} = \frac{0.05}{1 , 000 , 000} = 5 \times 1 0^{- 8}$

Yeh kyun matter karta hai: Yahi famous genome-wide significance threshold hai: p < 5×10⁻⁸. Isse kamzor koi bhi association likely ek false positive hai.

Odds Ratio

Effect size quantify karne ke liye, hum odds ratio (OR) calculate karte hain:

$O R = \frac{a / c}{b / d} = \frac{a d}{b c}$

Odds ratio kyun use karein?

OR = 1: Koi association nahin (allele frequency cases aur controls mein identical)
OR > 1: Risk allele (cases mein zyada common)
OR < 1: Protective allele (cases mein kam common)

First principles se derivation:

Allele A dene par disease ki odds: $\frac{a}{c}$ (cases with A : controls with A)
Allele G dene par disease ki odds: $\frac{b}{d}$
In odds ka ratio odds ratio deta hai

Chhote effect sizes ke liye (GWAS mein typical), hum often log odds ratio report karte hain kyunki yeh zero ke around symmetric hota hai aur distribution mein approximately normal hota hai.

[!example] Worked Example 1: Type 2 Diabetes SNP

Setup: Hum SNP rs7903146 ko TCF7L2 gene mein 5,000 Type 2 Diabetes (T2D) cases aur 5,000 controls mein genotype karte hain.

Data:

T2D cases: 3,200 ke paas T allele hai, 1,800 ke paas C allele hai
Controls: 2,400 ke paas T allele hai, 2,600 ke paas C allele hai

Step 1: Contingency table banao

	Cases	Controls	Total
T allele	3,200	2,400	5,600
C allele	1,800	2,600	4,400
Total	5,000	10,000

Yeh step kyun? Humein observed frequencies ko organize karna hai taaki independence ke under expected frequencies se compare kar sakein.

Step 2: Chi-square statistic calculate karo

$χ^{2} = \frac{10 , 000 ( 3 , 200 \times 2 , 600 - 2 , 400 \times 1 , 800 ) ^{2}}{5 , 600 \times 4 , 400 \times 5 , 000 \times 5 , 000}$

$= \frac{10 , 000 ( 8 , 320 , 000 - 4 , 320 , 000 ) ^{2}}{616 , 000 , 000 , 000 , 000}$

$= \frac{10 , 000 \times 16 , 000 , 000 , 000}{616 , 000 , 000 , 000} = 259.74$

Yeh step kyun? Bada $χ^{2}$ value null hypothesis se strong deviation indicate karta hai, matlab koi association nahin.

Step 3: P-value nikalo

$χ^{2} = 259.74$ aur df = 1 ke saath, humein p < 10⁻⁵⁷ milta hai (5×10⁻⁸ se bahut neeche).

Yeh step kyun? Yeh extremely small p-value genome-wide significance confirm karta hai — yeh SNP T2D se robustly associated hai.

Step 4: Odds ratio calculate karo

$O R = \frac{3 , 200 \times 2 , 600}{2 , 400 \times 1 , 800} = \frac{8 , 320 , 000}{4 , 320 , 000} = 1.93$

Interpretation: T allele carry karne wale individuals mein Type 2 Diabetes develop karne ki 1.93× zyada odds hoti hai C allele wale logon ke comparison mein.

Yeh step kyun? Jahan p-value batata hai ki association real hai, OR batata hai effect ki magnitude — odds mein 93% ki increase clinically meaningful hai.

[!example] Worked Example 2: Manhattan Plot Interpretation

Setup: Tumne schizophrenia ke liye GWAS run kiya hai 500,000 SNPs ke saath, 10,000 cases aur 10,000 controls mein.

Results visualization: Ek Manhattan plot y-axis par −log₁₀(p-value) aur x-axis par chromosomal position display karta hai.

Key features:

Genome-wide significance line: −log₁₀(5×10⁻⁸) = 7.3 par horizontal line
Line ke upar peaks: Significant associations
Chromosome clustering: Aas-paas kai SNPs association dikha rahe hain (linkage disequilibrium)

Example peak: Chromosome 6, position 25-35 Mb, jisme 15 SNPs p < 5×10⁻⁸ ke saath hain.

Yeh step kyun? Clustering batata hai ki yeh 15 independent findings nahin hain — yeh SNPs linkage disequilibrium (LD) mein hain, matlab yeh saath-saath inherit hote hain. Is region mein ek causal variant hai (ya kuch), aur baaki sirf use tag kar rahe hain.

Fine-mapping ke liye step: Causal variant identify karne ke liye:

Is region ke sabhi SNPs ke beech LD calculate karo
Conditional analysis karo: Har SNP ko top SNP ko control karte hue test karo
Jo SNPs significant rehte hain woh likely independent signals hain
Jo SNPs non-significant ho jaate hain woh sirf top SNP ko tag kar rahe the

Yeh kyun matter karta hai? Yeh genuine causal variants ko correlated bystanders se alag karta hai.

[!example] Worked Example 3: Power Calculation

Question: Humein kitne samples chahiye ek SNP detect karne ke liye jisme OR = 1.2 ho (chhota effect), genome-wide significance par (p < 5×10⁻⁸), 80% power ke saath?

Step 1: Parameters set karo

Effect size: OR = 1.2
Risk allele frequency: f = 0.3 (common variant)
Significance level: α = 5×10⁻⁸
Desired power: 1 - β = 0.80

Step 2: OR ko genetic relative risk mein convert karo

Ek multiplicative model ke liye: $GR = \frac{O R}{1 - f + f \times O R}$

Yeh formula kyun? OR case-control studies mein association measure karta hai, lekin underlying genetic model relative risk use karta hai. Yeh dono ke beech convert karta hai.

$GR R = \frac{1.2}{1 - 0.3 + 0.3 \times 1.2} = \frac{1.2}{1.06} = 1.132$

Step 3: Power formula apply karo

Chi-square test ke liye: $n = \frac{( Z _{α /2} + Z β ) ^{2}}{2 f ( 1 - f ) ( GR R - 1 ) ^{2} \times K \times ( 1 - K )}$

jahaan K = disease prevalence hai.

Yeh formula kyun? Statistical power depend karta hai:

Signal strength: $(GR R - 1)^{2}$ (effect size)
Genetic variance: $2 f (1 - f)$ (heterozygosity)
Sample composition: $K (1 - K)$ (case-control balance)

Schizophrenia ke liye, K ≈ 0.01 (1% prevalence), balanced design:

$n = \frac{( 5.45 + 0.84 ) ^{2}}{2 \times 0.3 \times 0.7 \times 0.13 2 ^{2} \times 0.01 \times 0.99}$

$= \frac{39.56}{0.0000714} \approx 554, 000 per group$

Interpretation: Humein ~1.1 million total samples chahiye (550K cases + 550K controls) ek common variant ko reliably detect karne ke liye jo chhota effect rakhta hai (OR = 1.2), genome-wide significance par.

Yeh kyun matter karta hai: Yahi explain karta hai kyun early GWAS (5,000-10,000 samples ke saath) zyaadatar common variant associations miss kar gaye. Current mega-GWAS consortia >500,000 samples ke saath zaroori hain, excessive nahin.

[!mistake] Common Mistakes

Mistake 1: "Significant SNP = Causal Gene"

Kyun sahi lagta hai: SNP strong association dikhaata hai, toh zaroor yahi disease cause kar raha hoga.

Kyun galat hai: Zyaadatar GWAS SNPs causal nahin hote — yeh true causal variant ke saath linkage disequilibrium mein hote hain. Significant SNP ek gene desert mein ho sakta hai, jabki causal variant 200 kb door ek regulatory region mein hai, alag gene ko affect karta hua.

Fix:

GWAS loci (regions) identify karta hai, causal variants nahin
Denser genotyping/sequencing ke saath fine-mapping chahiye
Mechanism identify karne ke liye functional follow-up (expression QTL, CRISPR validation) zaroori hai

Example: FTO locus obesity se associated tha, aur sabne maana ki FTO gene causal hai. Saalon baad, researchers ne discover kiya ki causal variants IRX3 aur IRX5 ko regulate karte hain, jo 500 kb door ke genes hain jo thermogenesis control karte hain.

Mistake 2: "No genome-wide significant hits = No genetic basis"

Kyun sahi lagta hai: Agar GWAS kuch significant nahin paata, toh trait genetic nahin hoga.

Kyun galat hai: Chaar possible explanations hain:

Insufficient power: Effect sizes tumhare sample size ke liye bahut chhote hain
Rare variants: GWAS chips common variants capture karte hain (MAF > 5%); rare variants ko sequencing chahiye
Structural variants: CNVs, inversions jo SNP arrays se capture nahin hote
Environmental interaction: Genetic effects sirf specific environments mein manifest hote hain (G×E)

Fix:

Twin/family studies se heritability estimates check karo (agar h² > 0 hai, toh genetic component HAI)
Sample size badhao (mega-consortia with 100K+ samples)
Rare variants ke liye sequencing-based approaches use karo
Weak signals aggregate karne wale polygenic risk scores (PRS) consider karo

Example: Schizophrenia mein h² ≈ 80% hai lekin early GWAS ne kuch nahin paaya. >100,000 samples ke saath, modern GWAS ne 200+ loci identify kiye, heritability ka ~20% explain karte hue. "Missing heritability" hazaaron variants mein distributed hai jo tiny effects rakhte hain.

Mistake 3: "P-value ranking = Effect size ranking"

Kyun sahi lagta hai: Sabse low p-value wala SNP zaroor sabse strong biological effect rakhta hoga.

Kyun galat hai: P-value effect size AUR sample size dono par depend karta hai (aur allele frequency par bhi). Ek common variant (MAF = 0.4) jisme OR = 1.15 hai, p = 10⁻⁵⁰ ho sakta hai, jabki ek rare variant (MAF = 0.01) jisme OR = 3.0 hai, p = 10⁻⁶ ho sakta hai.

Fix: Hamesha effect size (OR ya β) aur confidence intervals p-values ke saath report karo. Clinical/biological interpretation ke liye, effect size zyada matter karta hai.

Example: Height GWAS mein, top p-value SNP (p = 10⁻⁷⁰) height ka 0.4 cm explain karta hai, jabki ek rarer variant (p = 10⁻⁸) 2 cm explain karta hai. Baad wala zyada biologically interesting hai, kamzor p-value ke bawajood.

Mistake 4: "GWAS requires family studies or pedigrees"

Kyun sahi lagta hai: Genetics studies traditionally families use karti thin inheritance track karne ke liye.

Kyun galat hai: GWAS ek population-based approach hai jo unrelated individuals use karta hai. Power aati hai:

Large sample sizes se (deep pedigrees se nahin)
Common variant discovery se (rare Mendelian mutations nahin)
Cases aur controls ke beech population allele frequency differences se

Fix: Samjho ki GWAS aur linkage analysis complementary hain:

Linkage (family-based): Rare, high-penetrance variants ke liye achha hai (Mendelian diseases)
GWAS (population-based): Common, low-penetrance variants ke liye achha hai (complex traits)

Example: Huntington's disease (single CAG repeat expansion se caused) linkage analysis se map kiya gaya. Type 2 diabetes (polygenic, 500+ variants) GWAS se dissect kiya gaya.

[!recall]- Feynman Explanation (Age 12)

Socho tumhare school mein 1,000 bachhe hain, aur tum notice karte ho ki kuch bachhe bahut tall hain aur kuch short. Tum sochte ho: "Kya unke DNA mein kuch hai jo unhe tall banata hai?"

Tumhara DNA ek instruction book ki tarah hai jisme 3 billion letters hain (A, T, C, G). Kabhi kabhi ek bachhe ke paas position #12345 par "A" hota hai jabki doosre bachhe ke paas usi jagah "G" hota hai. Yeh tiny differences SNPs kehlate hain (isse "snips" bolo).

Scientists GWAS mein kya karte hain:

Woh sabki height measure karte hain
Woh har person mein 1 million in SNP positions ko check karte hain
Har SNP ke liye woh poochte hain: "Kya zyaadatar tall bachhe A rakhte hain, jabki zyaadatar short bachhe G rakhte hain?"

Agar unhe milta hai ki haan, tall bachhe position #12345 par A rakhne ke bahut zyada likely hain, toh woh SNP height se associated hai. Yeh ek clue milne jaisa hai!

Lekin tricky part yeh hai: Woh "A" hona tumhe akele tall nahin banata. Yeh sirf ek marker hai. Aise socho: Agar sabhi tall bachhe red backpacks pehnein, tum notice kar sakte ho "red backpack = tall bachha" correlation. Lekin backpack tumhe tall nahin banata! Isi tarah, SNP actually sirf us actual gene ke paas baitha ho sakta hai jo height affect karta hai.

Scientists ko bahut SAARA extra kaam karna padta hai yeh figure out karne ke liye ki kaun sa gene actually responsible hai aur kaise kaam karta hai. GWAS sirf unhe ek map deta hai "X marks the spot" ke saath jahan deeper dig karna hai.

Unhe itne saare log kyun chahiye? Kyunki har SNP sirf ek tiny, tiny difference karta hai — shayad 1 millimeter height ka. Itne chhote effect spot karne ke liye, tumhe hazaaron logon ki zaroorat hai. Yeh ek noisy room mein whisper sunne ki koshish ki tarah hai — tumhe kai logon ki zaroorat hai confirm karne ke liye "Haan, maine bhi suna!"

[!mnemonic] Memory Aids

GWAS = "Genome-Wide A-lot-of SNPs"

Genome-wide: Poora genome scan karo
Wide: Lakho se crore positions
Association: Correlation dhundhna, causation nahin
Studies: Population-based comparison

5×10⁻⁸ rule: "Five in a hundred million" harsh lagta hai, lekin yaad rakho: "Million tests, micro-p" (lakho SNPs test karne ke liye microscopic p-values chahiye).

OR interpretation:

OR = 1: "One is none" (koi effect nahin)
OR = 2: "Two times the trouble" (double the risk)
OR = 0.5: "Half the hazard" (protective)

LD (Linkage Disequilibrium): "Linked DNA goes Downstream together" — aas-paas ke SNPs blocks ke roop mein inherit hote hain, isliye woh correlate karte hain chahe sirf ek causal ho.

Key Concepts

Genome-Wide Association Study (GWAS): Poore genomes mein genetic variants ka systematic scan traits/diseases ke saath associations identify karne ke liye
Single Nucleotide Polymorphism (SNP): Ek specific genomic position par single-base DNA variation (jaise A vs. G)
Case-control design: Trait wale individuals (cases) aur bina trait wale (controls) ke beech allele frequencies compare karna
Genome-wide significance: P-value threshold 5×10⁻⁸ jo multiple testing correction account karta hai
Odds Ratio (OR): Effect size ka measure; risk allele dene par disease ki odds ka ratio vs. reference allele
Linkage Disequilibrium (LD): Chromosome par physical proximity ki wajah se alag loci par alleles ka non-random association
Manhattan plot: Visualization jo sabhi chromosomes mein −log₁₀(p-value) dikhata hai
Minor Allele Frequency (MAF): SNP par less common allele ki frequency (common variants: MAF > 5%)
Fine-mapping: Associated loci ke andar causal variants pinpoint karne ke liye follow-up analysis
Polygenic trait: Phenotype jo bahut saare genetic variants se influence hota hai, har ek ka chhota effect
Missing heritability: Families se estimated heritability aur discovered SNPs se explain ki gayi variance ke beech ka gap

Connections

Single Nucleotide Polymorphisms (SNPs): GWAS dwara interrogate kiye gaye genetic markers
Linkage Disequilibrium and Haplotypes: Explain karta hai kyun aas-paas kai SNPs association kyun dikhate hain
Polygenic Risk Scores: Individual disease risk predict karne ke liye GWAS results use karte hain
Heritability and Variance Components: GWAS findings trait heritability samajhne mein contribute karti hain
Expression QTL (eQTL) Studies: Mechanism identify karne ke liye GWAS ka functional follow-up
Mendelian Randomization: Causal inference ke liye GWAS variants ko instrumental variables ke roop mein use karta hai
Population Stratification: GWAS mein correction ki zaroorat wala confounding factor
Fine-mapping and Credible Sets: Causal variants narrow down karne ke liye statistical methods
Rare Variant Association Studies: Low-frequency alleles ke liye GWAS complement karte hain
Pharmacogenomics: Drug response traits par apply kiya gaya GWAS

Flashcards

Genome-wide association study (GWAS) kya hai? :: Ek observational study jo genome mein lakho genetic variants (SNPs) test karta hai taaki genotype aur phenotype ke beech associations identify ho sakein, cases aur controls ke beech allele frequencies compare karke.

Genome-wide significance threshold p < 5×10⁻⁸ kyun set hai?

Kyunki GWAS ~1 million independent SNPs test karta hai, Bonferroni correction ki zaroorat hai: 0.05 / 1,000,000 = 5×10⁻⁸ — family-wise error rate control karne ke liye aur multiple testing se false positives avoid karne ke liye.

GWAS mein odds ratio (OR) of 1.5 ka kya matlab hai?

Risk allele carry karne wale individuals mein disease/trait hone ki 1.5× (ya 50% zyada) odds hoti hai reference allele wale individuals ke comparison mein. OR = 1 matlab koi association nahin, OR > 1 matlab risk allele, OR < 1 matlab protective allele.

Linkage disequilibrium (LD) kya hai aur GWAS ke liye kyun matter karta hai?

LD alag loci par alleles ka non-random association hai same chromosome par physical proximity ki wajah se. GWAS mein, iska matlab hai ki ek significant SNP khud causal nahin ho sakta lekin paas mein true causal variant se correlated ho sakta hai, jiske liye actual causal variant identify karne ke liye fine-mapping chahiye.

Zyaadatar GWAS-identified SNPs ko turant causal kyun nahin declare kiya ja sakta?

Zyaadatar significant SNPs true causal variant ke saath linkage disequilibrium mein hote hain. Significant SNP non-coding region mein ya kisi alag gene mein bhi ho sakta hai, jo sirf paas mein located actual causal variant ko "tag" kar raha hota hai. Causation establish karne ke liye functional validation zaroori hai.

Manhattan plot kya hai aur significance line ke upar peaks kya indicate karte hain?

Manhattan plot sabhi chromosomes mein test kiye gaye sabhi SNPs ke liye −log₁₀(p-value) display karta hai. Genome-wide significance line (−log₁₀(5×10⁻⁸) = 7.3) se upar ke peaks genomic loci indicate karte hain jo trait se significantly associated hain. Ek peak mein kai aas-paas SNPs usually LD ki wajah se ek locus represent karte hain.

Agar 500,000 independent SNPs α = 0.05 ke saath test kar rahe ho toh genome-wide significance threshold calculate karo :: Bonferroni correction use karke: α_corrected = 0.05 / 500,000 = 1×10⁻⁷. Yeh p-value threshold hai jiske neeche associations statistically significant mani jaati hain multiple testing ke liye correct karne ke baad.

GWAS typically tens of thousands samples kyun chahiye?

Kyunki zyaadatar common variants ke chhote effect sizes hote hain (OR = 1.1-1.3). Chhote effects genome-wide significance par detect karne ke liye high statistical power chahiye, jo large sample sizes ki demand karti hai. Power scale hoti hai n ×2f(1-f)(OR-1)² ke saath, toh weak effects ko bahut saare samples chahiye.

GWAS aur linkage analysis mein kya farq hai? :: GWAS unrelated individuals use karta hai common variants (MAF > 5%) test karne ke liye population-based case-control designs mein. Linkage analysis families/pedigrees use karta hai rare, high-penetrance variants ki inheritance track karne ke liye. GWAS complex polygenic traits ke liye hai; linkage Mendelian diseases ke liye hai.

Agar GWAS mein koi genome-wide significant hits nahin milte, toh kya iska matlab hai trait ka koi genetic basis nahin hai?

Nahin. Null results ke possible reasons mein shamil hain: (1) chhote effects ke liye insufficient sample size/power, (2) causal variants rare hain (common SNP arrays se capture nahin hote), (3) structural variants jo interrogate nahin kiye gaye, (4) gene-environment interactions, (5) heterogeneity. Twin studies se heritability estimates GWAS null results ke bawajood genetic contribution confirm kar sakti hain.

GWAS ke context mein "missing heritability" kya hai?

Trait heritability jo family studies se estimate ki gayi (jaise schizophrenia ke liye h² = 80%) aur GWAS-discovered variants se explain ki gayi variance (jaise 20%) ke beech ka gap. Isse explain kiya jaata hai: bahut saare variants jinka effect detect karne ke liye bahut chhota hai, rare variants jo capture nahin hue, structural variants, epistasis, aur environmental interactions.

GWAS Manhattan plot mein 20 significant SNPs ke clustered peak ki interpretation kaise karein?

20 SNPs likely linkage disequilibrium mein hain, jo 20 independent findings ki jagah ek associated locus represent karte hain. Fine-mapping aur conditional analysis (har SNP ko top SNP control karte hue test karna) yeh determine kar sakta hai ki multiple independent signals hain ya sirf ek causal variant hai jo sabhi 20 SNPs ke dwara tag kiya ja raha hai.