Integrated multi-omics interpretation and alternative explanations

Practical 25: Genomics, systems biology and multi-omics

Learning objectives

Compare cross-layer evidence; examine measured common drivers; state which conclusions require independent or causal evidence.

Biological question and data

When two molecular layers associate with the same production response, what can be concluded?

Proteins, metabolites and response share a deliberately supplied latent factor. Diet and genotype also contribute. The measured features are not specified as direct causes of the response in the generator.

No biological download is required. Numerical inputs are hypothetical, not observed animal data. The runnable examples use base R. Count and abundance units are specified where used.

1. State the evidence at its measurement level

Align one protein, one metabolite and the response, all in the same animals.

Show R code 
helper_dir <- if (dir.exists("R")) "R" else "../R"
stopifnot(file.exists(file.path(helper_dir,"teaching_data.R")))
source(file.path(helper_dir,"teaching_data.R"))
d <- teaching_data()
meta <- d$metadata
z <- data.frame(protein=d$protein[1,],metabolite=d$metabolite[1,],response=meta$response,
 diet=meta$diet,batch=meta$batch,dosage=d$genotypes[meta$animal,"marker1"])
round(cor(z[,1:3]),2)
           protein metabolite response
protein       1.00       0.24     0.16
metabolite    0.24       1.00     0.21
response      0.16       0.21     1.00

Cross-layer agreement is an observed covariance pattern. It is not independent replication because the same animals and common driver contribute.

2. Account for supplied measured covariates

Compare raw correlation with correlation of residuals after adjustment for diet, batch and genotype dosage.

Show R code 
rp <- residuals(lm(protein~diet+batch+dosage,z))
rm <- residuals(lm(metabolite~diet+batch+dosage,z))
c(raw=cor(z$protein,z$metabolite),after_measured_adjustment=cor(rp,rm))
                      raw after_measured_adjustment 
                0.2367859                 0.2053569
Show R code 
stopifnot(length(rp)==48,length(rm)==48)

The unmeasured part of the latent factor remains shared. Adjustment can change the pattern without identifying a mechanistic direction.

3. Connect claims to a suitable next study

Distinguish prediction, response and mechanism before selecting follow-up.

Show R code 
follow_up <- data.frame(claim=c("predict new animals","diet changes a molecular profile","protein mediates performance"),
 next_step=c("validate full pipeline in eligible new animals",
 "appropriate allocated contrast with assay/batch checks",
 "temporal and perturbation evidence with explicit mediation assumptions"))
follow_up
                             claim
1              predict new animals
2 diet changes a molecular profile
3     protein mediates performance
                                                               next_step
1                         validate full pipeline in eligible new animals
2                 appropriate allocated contrast with assay/batch checks
3 temporal and perturbation evidence with explicit mediation assumptions

More omics layers may strengthen a description without resolving the key uncertainty. A targeted experiment can be more informative than another correlated panel.

4. Demonstrate an ambiguous equilibrium result

A final abundance can arise from different dynamic processes. Under constant production and first-order loss, equilibrium is production divided by loss.

Show R code 
mechanism <- data.frame(explanation=c('baseline','production_increase','loss_decrease'),
 production=c(4,8,4),loss=c(1,1,.5))
mechanism$equilibrium <- with(mechanism,production/loss)
mechanism
          explanation production loss equilibrium
1            baseline          4  1.0           4
2 production_increase          8  1.0           8
3       loss_decrease          4  0.5           8

Two different changes double the equilibrium. A single endpoint cannot distinguish them in this model. Time-resolved data or a suitable turnover measurement can provide additional information.

5. Build a competing-explanations table

For each cross-layer association, list a direct biological path, a common driver, technical structure and reverse direction where plausible. Then choose a follow-up whose expected outcomes differ between explanations.

A replication cohort addresses repeatability and transfer. A targeted perturbation addresses a proposed causal role. Better identifier or chemical evidence addresses what was measured. These follow-ups should be selected for the uncertainty present rather than treated as interchangeable confirmations.

Interpretation challenge

Why does agreement between RNA, protein and phenotype not provide three independent causal demonstrations?

The layers can share a treatment, cell-composition or technical driver. They are paired observations from the same units and may reflect correlated consequences rather than a demonstrated causal chain.

Interpretation and limits

Conditional correlations do not rule out unmeasured confounding, measurement error, reverse causation or selection. A mediation claim needs temporal ordering and justified assumptions, not simply a smaller diet coefficient after adding a molecular covariate.

Exercises

  1. Why is layer agreement not independent replication here?
  2. Does a residual correlation prove direct interaction?
  3. What should a null or inconsistent layer result contribute?
  1. The same animals and deliberately common generating driver supply both measurements.
  2. No; an unmeasured common cause can remain.
  3. It constrains the explanation and should be reported, with assay sensitivity and context considered.

Continue learning

B13 Association, B14 Networks, B15 Integration and B18 Synthesis.

Continue with research proposal. Return to the practical index or the unified map.