Logistic PCA explains differences between genome-scale metabolic models in terms of metabolic pathways

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Genome-scale metabolic models (GSMMs) provide comprehensive insights into cellular metabolism.
• Comparing diverse GSMMs is challenging due to limitations in current dimensionality reduction and clustering techniques, which often lack mechanistic interpretability and rely on subjective assumptions.

Purpose of the Study:

• To introduce a novel, mechanistically interpretable approach for clustering GSMMs.
• To develop a method that identifies specific reactions and pathways driving the separation of GSMMs.

Main Methods:

• Application of logistic principal component analysis (LPCA) for GSMM clustering.
• Analysis of diverse datasets including Escherichia strains, budding yeasts, human tissues, and Firmicutes strains.

Main Results:

• LPCA successfully clustered GSMMs, preserving microbial phylogenetic relationships and discerning human tissue-specific metabolic profiles.
• Performance was comparable to established methods like t-distributed stochastic neighborhood embedding (t-SNE) and Jaccard coefficients.
• Identified subsystems and reactions by LPCA align with existing biological knowledge, confirming its reliability.

Conclusions:

• LPCA offers an effective and reliable method for dissecting GSMMs and uncovering the mechanistic drivers of metabolic differences.
• The approach enhances the interpretability of GSMM comparisons across various biological contexts.