Explain omics integration (multi-omics) — Hinglish notes

6.5.5 · Biology › Systems Biology & Frontiers

Multi-omics integration HAI kya?

Central dogma layers ki natural ordering deta hai:

$DNA transcription RNA translation Protein catalysis Metabolite$

Har arrow regulated hota hai, isliye information har step par lose aur reshape hoti hai — aur yahi reason hai ki ek layer doosri layer ko perfectly predict nahi kar sakti.

Hume iska zaroorat kyun hai? (core justification)

Naïve view ka steel-man: "Bas genome sequence karo — woh master blueprint hai." Sahi lagta hai kyunki DNA hi baaki sab kuch cause karta hai. Lekin genome tumhare body ke har cell mein almost identical hota hai, phir bhi neuron ≠ muscle cell. Differences is baat mein hain ki kaunse genes express hote hain aur proteins kaise modify hoti hain — ye dynamic layers hain jo genome nahi dikhata.

Ye KAISE kiya jaata hai? (integration score scratch se derive karna)

Hume ek single number chahiye jo kahe: "kya do layers agree karti hain ki feature $X$ important hai?" Chalte hain ise build karte hain.

Step 1 — Har layer ko standardise karo. Layer $A$ (maan lo mRNA) aur layer $B$ (protein) ke alag-alag units hote hain. Har measurement ko z-score mein convert karo taaki compare kar sako:

$z_{i} = \frac{x _{i} - μ}{σ}$

Ye step kyun? Kyunki tum "RNA ke counts" ko seedha "μg of protein" mein add nahi kar sakte — z-scoring dono ko unitless "ye value kitni unusual hai" numbers bana deta hai. Khaas baat ye hai ki proper z-scores ki mean 0 aur variance 1 hoti hai, jo final correlation ko $[- 1, 1]$ ke andar rakhti hai.

Step 2 — $n$ samples mein agreement measure karo. Ek gene jo dono layers mein measure hua ho, uska Pearson correlation z-scores ke products ka average hota hai:

$r_{A B} = \frac{1}{n - 1} \sum_{i = 1}^{n} z_{i}^{A} z_{i}^{B}$

Ye step kyun? Agar mRNA aur protein samples mein saath upar-neeche aate hain, toh unke z-scores ka sign ek jaisa hoga, products positive honge, aur $r \to + 1$ : strong cross-omics support.

Step 3 — Variation ke shared axes dhundo (matrices kyun). Layers ko ek badi data matrix mein stack karo aur woh directions ("latent factors") dhundo jo jointly vary karte hain. Singular Value Decomposition yahi karta hai:

$M = U Σ V^{T}$

Ye step kyun? $V$ ke columns batate hain ki genes/proteins ka kaunsa combination saath chalata hai. Sabse bada singular value $σ_{1}$ omics mein sabse strong shared signal hai — yahi MOFA aur iCluster jaise tools ka mathematical heart hai.

Figure — Explain omics integration (multi-omics)

Integration ki do strategies

Worked examples

Common mistakes

Forecast-then-Verify

Flashcards

Central-dogma order mein chaar canonical omics layers kaun si hain?

Genome (DNA) → Transcriptome (RNA) → Proteome (protein) → Metabolome (metabolites).

Transcriptomics akele protein levels jaanne ke liye kyun kaafi nahi?

mRNA–protein correlation sirf ~0.4 hai, toh protein variance ka ~84% translation/degradation/modification se aata hai jo RNA mein capture nahi hota.

Multi-omics integration define karo.

Computationally ≥2 omics layers ko same samples se combine karna taaki aisa pattern mile jo kisi ek akele layer mein visible nahi hota.

Do z-scored layers ke beech cross-omics agreement ka formula?

r_{A B} = \frac{1}{n - 1} \sum z_{i}^{A} z_{i}^{B}

, range −1 to 1.

r compute karne se pehle standardised vectors ki unit variance kyun honi chahiye?

Warna

\frac{1}{n - 1} \sum z_{i}^{A} z_{i}^{B}

1 se zyada ho sakta hai, ek impossible correlation deta hai; unit variance r ko [−1,1] mein rakhti hai.

Early aur late integration mein kya fark hai?

Early = raw layers ko concatenate karo phir ek baar analyse karo; Late = har layer ko alag analyse karo phir results combine karo.

Early integration mein concatenate karne se pehle har layer ke saath kya karna chahiye?

Standardise/z-score karo (mean 0, variance 1) taaki koi bhi layer scale ya feature count se dominate na kare.

Stacked z-scored omics ka top singular vector kya represent karta hai?

Layers mein shared variation ka sabse strong axis (ek latent factor).

Ek gene ka mRNA high hai lekin protein low — biological interpretation?

Post-transcriptional regulation: reduced translation ya rapid protein degradation.

Jab r=0.4 ho toh shared nahi hone wala variance fraction kitna hai?

1 - r^{2} = 0.84

(84%).

Genome akele cell-type differences kyun explain nahi kar sakta?

Genome cells mein ~identical hota hai; differences is baat mein aate hain ki kaunse genes express hote hain aur proteins kaise modify hoti hain (dynamic layers).

Recall Feynman: 12-saal ke bachche ko explain karo

Socho ek school play hai. Script (genome) har performance ke liye same hoti hai. Lekin har raat actors alag-alag lines zor se ya dheere bolte hain (RNA), kuch actors beemar pad jaate hain aur replace ho jaate hain (proteins), aur audience alag react karti hai (metabolites). Agar tum sirf script padho, tum kabhie nahi jaanoge ki kal raat ka show flop kyun hua. Multi-omics matlab hai sab kuch dekhna — script, actors, aur audience — taaki aakhirkar samajh aaye stage par asal mein kya hua.

Connections

Central Dogma of Molecular Biology — omics layers ki ordering
Gene Expression Regulation — kyun RNA ≠ protein
Principal Component Analysis & Singular Value Decomposition — shared latent factors ki math
Pearson Correlation — cross-layer agreement metric
Precision Medicine — clinical payoff: multi-omics tumour subtyping
Systems Biology — parent framework jo cell ko ek integrated network maanta hai