Questions: Gene Regulatory Networks

Question 1

Why does gene co-expression not necessarily imply a direct regulatory relationship?

• A. Because co-expression analysis uses the wrong statistical test
• B. Because two genes may be co-expressed due to shared upstream regulation, shared response to the same signal, or indirect regulatory cascades
• C. Because co-expression can only be detected in single-cell data
• D. Because transcription factors do not affect other genes' expression

Answer: (B)

If genes A and B are both activated by transcription factor C in response to a stimulus, they will show correlated expression across conditions — but neither regulates the other. Co-expression networks capture this coordinated behavior and are useful for identifying functional modules, but the edges represent statistical association, not direct regulation. Establishing direct regulation requires additional evidence: ChIP-seq showing the transcription factor binds the target gene's regulatory region, or perturbation experiments showing that changing the regulator's activity directly changes the target's expression.

Question 2

Bayesian network inference from gene expression data can determine the complete, true gene regulatory network of a cell.

• T. True
• F. False

Answer: False

Bayesian network inference from observational expression data faces fundamental limitations: it cannot distinguish between correlation and causation, many different network structures can explain the same data equally well (non-identifiability), and the number of possible networks grows super-exponentially with the number of genes. Furthermore, regulatory networks are context-dependent (different in different cell types and conditions), dynamic (changing over time), and operate at multiple levels (transcriptional, post-transcriptional, protein-level) — no single inference method captures all of this. GRN inference produces plausible hypotheses that require experimental validation.

Question 3

Explain how integrating ChIP-seq data with RNA-seq perturbation data strengthens gene regulatory network inference compared to using either data type alone.

Model answer: ChIP-seq identifies where a transcription factor physically binds in the genome (direct targets), but binding does not guarantee functional regulation — many binding events have no measurable effect on gene expression. RNA-seq after perturbing (knocking out or overexpressing) the transcription factor identifies genes whose expression changes in response, establishing functional relevance, but cannot distinguish direct from indirect targets. Combining both reveals genes that are both directly bound by the transcription factor and change expression when it is perturbed — the most confident set of direct, functional regulatory targets. This integration eliminates false positives from both individual approaches.

This is the gold standard for GRN edge validation: direct binding evidence (ChIP-seq) plus functional consequence (perturbation RNA-seq). Large-scale efforts like ENCODE have generated both data types for hundreds of transcription factors, enabling comprehensive GRN reconstruction for well-studied cell types.