Multi-omics

Multiomics integration

Integrate multiple omics layers: microbiome, metabolome, transcriptome, etc.
Explore associations and predict outcomes
Gain complementary insights that single-omics cannot provide

Data containers

altExp() slot in TreeSummarizedExperiment
MultiAssaySummarizedExperiment

Methods

Association: detect cross-omics correlations
Ordination: reduce dimensionality, e.g., MOFA2
Supervised machine learning: predict outcomes, e.g., IntegratedLearner

Cross-association

CLR + Spearman

Pearson correlation

Measures linear association between two features
Spearman correlation = Pearson correlation on ranks

\[ r = \frac{ \underbrace{\sum_{i=1}^{n} (x_i - \bar{x})(y_i - \bar{y})}_{\text{numerator: covariance}} }{ \underbrace{\sqrt{\sum_{i=1}^{n} (x_i - \bar{x})^2} \;\; \sqrt{\sum_{i=1}^{n} (y_i - \bar{y})^2}}_{\text{denominator: product of standard deviations}} } \]

Calculating “targeted” correlations

Focus on feature pairs with found relationships
Avoids computing all pairwise correlations
Implemented in anansi

Ordination

Multi-omics Factor Analysis (MOFA2)

Factorizes multiomics matrix into latent factors
Captures shared and data-specific variation across omics
Outputs low-dimensional embeddings for samples and features

Supervised ML

IntegratedLearner

Predict phenotypes from multiple omics
Ensemble learning combining base learners per omics

Demonstration

library(mia)
library(pheatmap)

# Calculate correlations
mat <- getCrossAssociation(
  mae, assay.type1 = "rclr", experiment2 = 3, 
  assay.type2 = "signals", mode = "matrix"
)

library(mia)
library(pheatmap)

# Calculate correlations
mat <- getCrossAssociation(
  mae, assay.type1 = "rclr", experiment2 = 3, 
  assay.type2 = "signals", mode = "matrix"
)
# Visualize
pheatmap(mat)

Exercises

From OMA online book, Chapter 23: Cross-association

All exercises