6.4.11
Genomic data itna bada aur itna abstract hota hai ki human eye ke liye ise raw numbers ke roop mein samajhna impossible hai. Ek akele genome mein ~3 billion bases hote hain; ek RNA-seq experiment genes × samples ki ek matrix produce karta hai jisme tens of thousands of genes aur hundreds of samples hote hain. Hamari aankhein patterns detect karti hain, numbers nahi padhti. Data visualization invisible statistical structure ko visible spatial structure mein convert karta hai taaki hum ek peak, ek cluster, ya ek outlier dekh sakein jo kisi bhi table mein nazar nahi aata.
Genomic data visualization ek aisi practice hai jisme quantitative aur positional biological data (sequence coordinates, read counts, expression levels, variant positions) ko visual channels (position, length, color, angle) pe map kiya jaata hai taaki biological patterns perceivable ho jayein.
Core idea hai visual encoding : har data piece ko ek aisi channel assign ki jaati hai jise aankhein achhi tarah se padh sakti hain.
Data type
Best visual channel
Example plot
Genomic position
horizontal position (x-axis coordinate)
genome browser track
Read count / coverage
height (bar length)
coverage histogram
Expression value
color intensity
heatmap
Sample similarity
2D distance
PCA /clustering
Genome-wide signal
angle around a circle
Circos plot
Intuition Channels kyun important hainAankhein position ko sabse accurately judge karti hain, uske baad length ko, aur color ko sabse kam accurately. Isliye hum sabse important precise variable ko position pe rakhte hain, aur color ka use sirf broad patterns dikhane ke liye karte hain — kabhi exact values padhne ke liye nahi.
Woh 20% jo 80% value deta hai : teen "workhorse" plot families.
Genome browser tracks — answer karta hai "chromosome pe kahan?" Heatmaps + clustering — answer karta hai "kaunse genes/samples ek jaisa behave karte hain?" Manhattan & volcano plots — answer karta hai "kaunse points statistically significant hain?"
Agar tum ye teeno master kar lo, tum zyaadatar genomics papers padh sakte ho.
Ek chromosome one-dimensional hota hai — coordinates ki ek line. Isliye natural layout yeh hai: coordinate ko x-axis pe rakho, alag-alag data ("tracks") ko vertically stack karo. Har track ek hi x share karta hai isliye tum features ko align kar sakte ho.
Coverage track : har base position p C ( p ) = ∑ r 1 [ r overlaps p ] Gene track : boxes (exons) jo thin lines (introns) se jude hote hain.Variant track : SNP positions pe tick marks.
Ek heatmap ek matrix M M ij i j color scale ke zariye.
Standard preprocessing derive karna — the Z-score.
Raw expression values bohot bade ranges mein hote hain (housekeeping genes >> rare genes), isliye unka color sab kuch dabaa dega. Hum chahte hain ki har gene ko apne baseline se compare kiya jaaye. Gene i
z ij = σ i x ij − μ i
μ i σ i + 2 har gene ke liye.
μ i = n 1 ∑ j = 1 n x ij , σ i = n 1 ∑ j = 1 n ( x ij − μ i ) 2
Phir rows aur columns ko hierarchical clustering se reorder kiya jaata hai taaki similar profiles adjacent ho jayein, jo co-regulated genes ke blocks ko visible banata hai.
"Interesting" differentially-expressed genes dhundhne ke liye hume dono chahiye — bada change AUR statistical confidence . Ek axis dono nahi dikha sakta — isliye do use karo.
x-axis : change ki magnitude = log 2 ( fold change ) × 2 × 2 1 log 2 2 = + 1 log 2 2 1 = − 1 y-axis : confidence = − log 10 ( p -value ) − log 10 1 0 − 8 badi heights (8
x = log 2 expr control expr treat , y = − log 10 ( p )
Upar ke do "wings" = genes jo strongly aur significantly dono badal gaye hain.
Wahi y = − log 10 ( p )
Worked example Example 1 — Heatmap cell ke liye Z-scoreGene A ki expression 4 samples mein: [ 10 , 12 , 8 , 10 ] 12 z
Step 1 : μ = ( 10 + 12 + 8 + 10 ) /4 = 10 Kyun? Gene ko apne mean pe center karo.Step 2 : deviations [ 0 , 2 , − 2 , 0 ] σ = ( 0 + 4 + 4 + 0 ) /4 = 2 ≈ 1.41 Kyun? Spread se scale karo.Step 3 : z = ( 12 − 10 ) /1.41 = + 1.41 Kyun? Sample 2, average se 1.41 SD upar hai → warm/red color mein.
Worked example Example 2 — Volcano coordinatesEk gene: control mean = 50, treated mean = 200, p = 1 0 − 6
Step 1 : fold change = 200/50 = 4 Ratio kyun, difference kyun nahi? Biology multiplicative hoti hai.Step 2 : x = log 2 4 = 2 Kyun? 4-fold upar → x = +2.Step 3 : y = − log 10 ( 1 0 − 6 ) = 6 Kyun? Bohot significant → upar.Result : point ( 2 , 6 )
Worked example Example 3 — Coverage track padhnaReads in positions cover karte hain: base 100 ko 5 reads, base 101 ko 8, base 102 ko 3.
Track x=100,101,102 pe height 5, 8, 3 ki bars dikhata hai. 102 pe ek dip ka matlab ho sakta hai poor mapping ya deletion. Kyun dekhte hain? Achanak coverage drops structural variants flag karte hain.
Common mistake "Red ka matlab hamesha high expression hai"Kyun sahi lagta hai: hum intuitively padhte hain red = hot = bada.
The catch: Z-score heatmap mein, red ka matlab hai "is gene ke mean ke relative zyaada," na ki high absolute expression. Value 5 wala ek rare gene red ho sakta hai jabki value 5000 wala housekeeping gene blue ho.
Fix: Hamesha check karo ki heatmap raw, log, ya Z-scored data pe hai — color-scale legend padho.
Common mistake "Volcano plot mein sabse uncha point sabse important gene hai"Kyun sahi lagta hai: tall = most significant.
The catch: Height = statistical confidence, effect size nahi. Tiny fold change lekin huge sample size wala gene bohot tall ho sakta hai phir bhi biologically trivial ho.
Fix: Gene ke liye require karo ki woh dono thresholds pass kare — x pe dur aur y pe upar (top corners).
Common mistake "Rainbow color scale use karna theek hai"Kyun sahi lagta hai: rainbows colorful aur distinct hote hain.
The catch: Rainbow scales perceptually uniform nahi hote — equal numeric steps unequal color jumps jaisi dikhti hain, fake boundaries create karti hain, aur color-blind readers ke liye fail hoti hain.
Fix: Perceptually uniform / diverging scales use karo (jaise viridis, Z-scores ke liye blue-white-red).
Recall Feynman: 12-year-old ko explain karoSocho tumhare paas ek giant spreadsheet hai jisme tumhare body ke genes ke baare mein ek million numbers hain. Unhe ek ek karke padhna impossible hai. Toh instead tum use paint karo: har number ek color ban jaata hai, bade ke liye hot aur chhote ke liye cold. Ab spreadsheet ek picture jaisi dikhti hai, aur tumhari aankhein instantly red splotches spot kar leti hain — "interesting" genes. Doosre tricks DNA ke ek map pe dots rakhte hain (jahan cheezein hoti hain) ya dots is tarah plot karte hain ki sach mein important wale screen ke top pe ud jaayein. Visualization = boring numbers ko aisi pictures mein badalna jo tumhari aankhein padh sakein.
"Position, Length, Color — eye ki Precision fades." (Trust ka order: exact data ko position pe rakho, patterns ko color pe.)
Plot choice ke liye: "HMV" — H eatmap (kaun alike hai), M anhattan (kahan), V olcano (kya badla hai).
Heatmaps mein raw expression ki jagah Z-scores kyun use hote hain? Har gene ko uske apne mean pe center karne ke liye aur uske apne SD se scale karne ke liye, taaki sab genes comparable ho jayein aur ek high-expression gene color dominate na kare.
Volcano plot ka x-axis kya represent karta hai aur yeh log kyun hai? log 2
Volcano aur Manhattan dono plots ka y-axis kya dikhata hai? − log 10 ( p -value )
Human eye sabse accurately kaunsa visual channel padhti hai? Position, uske baad length, phir (sabse kam accurately) color.
Genome browser mein coverage track kya plot karta hai? x-axis coordinate ke saath har base position ke upar overlapping reads ki sankhya.
Rainbow color scales se kyun bachna chahiye? Yeh perceptually uniform nahi hote (equal steps unequal dikhte hain, false edges create karte hain) aur color-blind viewers ke liye bure hote hain.
Volcano plot pe "interesting" genes kaunse hote hain? Jo x=0 se dur hote hain (bada change) AUR y pe upar hote hain (significant) — dono top corners.
Gene i j z ij = ( x ij − μ i ) / σ i
RNA-seq analysis — expression matrices produce karta hai jo heatmaps banti hain.GWAS — Manhattan plots ka source.Hierarchical clustering — heatmap rows/columns ko reorder karta hai.Principal Component Analysis — sample similarity ki 2D visualization.Log transformations — kyun log 2 − log 10 Genome browsers (IGV, UCSC) — track-based visualization tools.Multiple testing correction — Manhattan/volcano plots pe significance line set karta hai.
