RNA-seq shows that a metabolic enzyme's mRNA is upregulated 5-fold in cancer cells compared to normal cells. Can you conclude the metabolic pathway is 5-fold more active?
BNo — mRNA levels are a poor proxy for flux because flux depends on protein abundance (which may not track mRNA due to translational regulation), post-translational modifications (which may activate or inhibit the enzyme), substrate availability, allosteric regulation, and thermodynamic constraints
CYes — gene expression is the only determinant of metabolic activity
DNo — but only because cancer cells have defective ribosomes
This is the central motivation for multi-omics integration. The correlation between mRNA and protein levels is notoriously imperfect (R^2 ~ 0.4-0.6 in many studies). Even protein abundance does not determine enzyme activity, because post-translational modifications (phosphorylation, acetylation), allosteric regulators, and substrate/product concentrations all modulate flux independently of enzyme quantity. Understanding metabolic flux requires integrating transcriptomic, proteomic, metabolomic, and potentially fluxomic data — each layer adds constraints that the others cannot provide.
Question 2 True / False
The simplest multi-omics integration approach — concatenating all data into one large matrix and applying standard machine learning — always outperforms single-omics analysis.
TTrue
FFalse
Answer: False
Concatenation (sometimes called 'early integration') often suffers from the curse of dimensionality: combining thousands of features from multiple omics layers creates a very high-dimensional space where signal can be diluted by noise, especially when sample sizes are small. Different omics layers may have different noise properties, different scales, and different missing data patterns. Naive concatenation can actually perform worse than analysis of the most informative single layer. More sophisticated approaches (MOFA, network-based integration, kernel methods, mechanistic model-based integration) handle these challenges by respecting the structure of each data type and the biological relationships between layers.
Question 3 Short Answer
Describe how multi-omics data can be integrated with genome-scale metabolic models to improve flux predictions.
Think about your answer, then reveal below.
Model answer: Transcriptomic data constrains reaction flux bounds in FBA: if an enzyme's mRNA is absent, its reaction flux is set to zero; if highly expressed, the upper bound may be increased. Proteomic data provides better flux constraints because protein abundance more directly reflects catalytic capacity. Metabolomic data constrains the thermodynamic feasibility of reactions (reactions must proceed in the direction of negative Gibbs free energy, which depends on metabolite concentrations). Methods like GIMME, iMAT, and E-Flux map expression data onto the metabolic model, and thermodynamic approaches (like TMFA) use metabolomics to add thermodynamic constraints. Each omics layer adds constraints that narrow the feasible flux space, improving predictions beyond what stoichiometry and a biomass objective alone can achieve.
The progression from unconstrained FBA (stoichiometry only) to expression-constrained FBA to thermodynamically constrained FBA to fully multi-omics-constrained models represents increasing biological realism. Each additional data layer reduces the feasible flux space, bringing predictions closer to the cell's actual metabolic state.