Describe variant calling pipelines

6.4.10 · Biology › Bioinformatics & Computational Biology

HOW: the pipeline stages

Figure — Describe variant calling pipelines

Canonical DNA-seq pipeline kuch is tarah flow karta hai: Raw reads → QC/trim → Align → Post-process → Call → Filter → Annotate.

1. Quality control & trimming

WHY: Sequencers har base ke liye ek Phred quality score attach karte hain. Low-quality ends aur adapter contamination aage jaake false variants create karte hain.
HOW: Pehle ek QC-reporting tool se quality assess karo (FastQC — ye sirf reports banata hai, ye trim nahi karta). Phir dedicated trimming tools se adapters aur low-quality bases trim karo (Trimmomatic, fastp).

2. Alignment (mapping)

WHY: Hume pata hona chahiye ki har read kahan se aayi thi, tabhi hum usse reference se compare kar sakte hain.
HOW: Burrows–Wheeler-based aligner se reads ko reference ke saath align karo (BWA-MEM, Bowtie2). Output = SAM/BAM file. Har read ko ek MAPQ mapping-quality score milta hai (same Phred logic: probability ki read galat jagah placed hai).

3. Post-alignment processing

BAM ko coordinate ke hisaab se Sort & index karo.
Mark duplicates — WHY: PCR/optical duplicates ek hi original molecule ki copies hain; unhe independent evidence ki tarah count karna ek variant ke liye fake support create karta hai. Hum unhe flag karte hain taaki woh ek baar count hon.
Base Quality Score Recalibration (BQSR) — WHY: Sequencers systematically $Q$ ko galat estimate karte hain; recalibration numbers ko correct karta hai taaki probability model sahi ho.

4. Variant calling proper — the Bayesian core

Formula Genotype likelihood via Bayes — from scratch

Goal: Sabse zyada probable genotype $G$ (jaise AA, AG, GG) dhundho given observed reads $D$ . Step 1 — Bayes' theorem (WHY: hum sirf $P (data ∣ genotype)$ compute kar sakte hain error model se, lekin hum chahte hain $P (genotype ∣ data)$ ): $P (G ∣ D) = \frac{P ( D ∣ G ) P ( G )}{P ( D )}$

Why this step? $P (D ∣ G)$ base error rates se computable hai; prior $P (G)$ population allele frequencies se aata hai. Step 2 — likelihood of the data. Reads (assumed) independent hain, isliye per-read probabilities multiply karo: $P (D ∣ G) = \prod_{i = 1}^{n} P (d_{i} ∣ G)$
Why? Independence ek mushkil joint probability ko ek simple product mein badal deta hai. Step 3 — single read under a heterozygote (genotype mein do alleles $b_{1}, b_{2}$ hain, read base $d_{i}$ show karta hai error prob $ϵ_{i} = 1 0^{- Q_{i} /10}$ ke saath): $P (d_{i} ∣ G = b_{1} b_{2}) = \frac{1}{2} P (d_{i} ∣ b_{1}) + \frac{1}{2} P (d_{i} ∣ b_{2})$ jahan $P (d_{i} ∣ b) = 1 - ϵ_{i}$ agar $d_{i} = b$ , warna $ϵ_{i} /3$ .
Why the ½? Do alleles mein se har ek equally likely hai ki wo kisi given read ke liye template ho.
Why $ϵ /3$ ? Agar error hai, to ye 3 baaki bases mein se koi bhi ho sakta hai, equally split. Step 5 — decide. Wo $G$ choose karo jo $P (G ∣ D)$ maximize kare; confidence ko Phred-scaled QUAL score ke roop mein report karo. Tools: GATK HaplotypeCaller, FreeBayes, bcftools, DeepVariant (CNN-based).

Output = ek VCF (Variant Call Format) file: har variant ke liye ek line jisme CHROM, POS, REF, ALT, QUAL, aur per-sample genotypes hain.

5. Filtering

WHY: Achhe callers bhi galtiyan karte hain. Low QUAL, low depth (DP), strand bias, etc. wale variants ko remove karo. (Hard filters ya GATK ki VQSR machine-learning recalibration.)

6. Annotation

WHY: Ek position meaningless hai jab tak pata na ho ki wo kaunse gene/effect ko hit karta hai.
HOW: ANNOVAR, VEP, SnpEff jaise tools variants ko label karte hain (missense, nonsense, synonymous, coding vs intron), aur dbSNP, ClinVar, gnomAD se cross-reference karte hain.

Worked examples

Common mistakes

Recall Feynman: explain to a 12-year-old

Tumne ek badi book ko confetti mein faad ke har tukde ko scan kiya. Typos dhundne ke liye, pehle pata lagao ki har tukda original book mein kahan belongs karta hai (alignment). Phir, har jagah par, un saare tukdon ko dekho jo usse cover kar rahe hain. Agar zyaatar tukde agree karte hain ki letter badal gaya, to ye ek real typo hai (variant). Agar sirf ek weird tukda disagree karta hai, to shayad wo ek smudge hai (sequencing error). Tum un tukdon ko bhi fenko dete ho jo simply usi page ki Xerox-copies hain taaki tum soch lo ki smudge real hai. Yahi variant calling hai!

Flashcards

Variant calling kya identify karta hai?

Wo positions jahan sample ka genome reference genome se alag hota hai, SNVs, indels, ya structural variants ke roop mein classify kiya gaya.

Phred score formula aur uska inverse define karo.

Q = - 10 lo g_{10} P

aur

P = 1 0^{- Q /10}

, jahan P base-call error probability hai.

Q30 ke corresponding error probability kya hai?

1 0^{- 3} = 0.001

(1 in 1000).

Kya FastQC trimming karta hai?

Nahi — FastQC sirf quality-control reports generate karta hai; trimming Trimmomatic ya fastp jaise tools karte hain.

Alignment ka standard output file kya hai?

SAM/BAM (BAM = binary compressed SAM).

MAPQ kya hai?

Mapping quality: Phred-scaled probability ki ek read galat location par aligned hai.

Calling se pehle PCR duplicates mark kyun karte hain?

Duplicates ek hi original molecule ki copies hain; unhe independent evidence ki tarah treat karna genotype likelihood product mein independence assumption ko violate karta hai aur false confidence inflate karta hai.

Genotype calling ke liye Bayes' theorem state karo.

P (G ∣ D) = P (D ∣ G) P (G) / P (D)

; wo G choose karo jo posterior maximize kare.

P (D ∣ G)

reads par product kyun hai?

Reads ko independent assume kiya jaata hai, isliye joint probability = per-read probabilities ka product.

Ek heterozygote AG ke under, P(read shows A) kya hai?

\frac{1}{2} (1 - ϵ) + \frac{1}{2} (ϵ /3) \approx 0.49967 \approx 0.5

, jahan

ϵ

base error probability hai.

Called variants kaun se file format mein store hote hain?

VCF (Variant Call Format).

BQSR kya correct karta hai?

Sequencer dwara base quality scores ki systematic miscalibration.

Pipeline stages ka order batao.

QC/trim → Align → Post-process (sort, mark duplicates, BQSR) → Call → Filter → Annotate.

Variants ko annotate kyun karte hain?

Biological meaning assign karne ke liye (gene, effect: missense/nonsense/synonymous) aur dbSNP/ClinVar/gnomAD jaise databases se cross-reference karne ke liye.

Connections

Phred Quality Scores
Sequence Alignment (BWA / Burrows-Wheeler Transform)
Bayes' Theorem
VCF and BAM File Formats
SNPs and Genetic Variation
GATK Best Practices
Next-Generation Sequencing
Genome Annotation

WHAT is variant calling?