Fig 1.
Schematic workflow of applying LPCA to binary reaction matrices derived from GSMMs.
(Created with BioRender.com).
Fig 2.
LPCA (a), t-SNE (b) and Jaccard coefficients (c) derived from a binary reaction matrix from differential reactions in 222 Escherichia GSMMs.
In panels (a) and (b), points represent individual GSMMs, with different genera indicated by distinct symbols and colors. The top row in panel (c) uses these same colors to indicate the corresponding genera. Circles in panel (a) highlight clusters of E. albertii strains (blue), E. fergusonii strains (red), and a mixed cluster of E. coli, S. dysenteriae, S. flexneri, and Clades II to VIII (orange). Labeled arrows in panel (a) denote subsystem-centric loading vectors from LPCA (refer to the results and methods section for definitions).
Fig 3.
Impact of subsystems derived from LPCA and MLR for Escherichia GSMMs.
MLR: contribution of subsystems to phylogenetic classification normalized to the maximum value. LPCA: subsystem-centric loadings normalized to maximum loading (refer to methods section for details).
Fig 4.
LPCA (a), t-SNE (b), and Jaccard coefficients (c) derived from a binary reaction matrix from yeast-specific GSMMs.
In panels (a) and (b), points represent individual GSMMs, with different genera indicated by distinct symbols and colors. The top row in panel (c) uses these same colors to indicate the corresponding genera. Circles in panel (a) highlight a cluster of the Lipomycetaceae clade (purple), and the Saccharomycodaceae (green). Labeled arrows in panel (a) denote subsystem-centric loading vectors from LPCA (refer to the results and methods section for definitions).
Fig 5.
LPCA (a), t-SNE (b), and Jaccard coefficients (c) derived from a binary reaction matrix from context-specific reconstructions from healthy and cancerous tissues.
In panels (a) and (b), points represent individual reconstructions, with reconstructions from healthy (blue squares) and cancerous tissues (red triangles). The top row in panel (c) uses these same colors to indicate the corresponding tissue. t-SNE appears less effective in identifying outliers (red open triangle, and blue open circles) compared to LPCA and Jaccard coefficients. Labeled arrows in panel (a) denote subsystem-centric loading vectors from LPCA (refer to the results and methods section for definitions).
Fig 6.
Impact of subsystems from LPCA and MLR for human reconstructions.
MLR: contribution of subsystems to phylogenetic classification normalized to the maximum value. LPCA: subsystem-centric loadings normalized to maximum loading (refer to methods section for details).
Fig 7.
LPCA (a), t-SNE (b) and Jaccard coefficients (c) derived from a binary reaction matrix from 2943 Firmicutes species-GSMMs.
In panels (a) and (b), points represent individual GSMMs, with different genera indicated by distinct symbols and colors. The top row in panel (c) uses these same colors to indicate the corresponding genera. The clustering of rows and columns in panel (c) was performed using the default hierarchical clustering settings (refer to methods section for details).