Extended Data Fig. 5: Statistical properties of LASSO analysis in MPCA.

(a) Evaluation of LASSO associations with global FDR q < 0.05. P-values were calculated by performing ordinary least squares (OLS) on those selected variables from LASSO, and p-values were computed from the OLS model using a two-sided t-test. FDR values were calculated by adjusting P-values with the Benjamini-Hochberg procedure. FDR q < 0.05 was considered significant. (b) Number and percentage of LASSO associations in BAT and liver that reached global FDR q < 0.05. (c) Enrichment over random selection of significant LASSO associations in BAT and liver in recapitulating physical interactions between proteins and metabolites in RHEA and TCDB. (d) Top1 LASSO predictors of metabolites with literature evidence in liver. (e) Top1 LASSO predictors of metabolites with literature evidence in BAT. (f) Extreme outliers in LASSO analysis identified in BAT. (g) Extreme outliers in LASSO analysis identified in liver. (h) LASSO Coefficient of the CDO1-hypotaurine edge. (i) LASSO Coefficient of the TYMP-thymine edge. (j) LASSO Coefficient of the PCY2-CTP-ethanolamine edge. (k) Percent of extreme outliers in BAT and liver with literature evidence. (l) Extreme outlier edges involving CML1, ACY3, and acetylated amino acids. (m) Validation score of metabolites in MPCA. For each metabolite, LASSO edges with literature evidence in RHEA, TCDB, and Reactome were counted. This was then used to linearly scale, in each tissue, to a score from 1–10 based on the number of recapitulated LASSO edges all the other metabolites have. The score from both tissues were then summed up to produce an overall validation score for each metabolite. (n) Annotation of putative function of LASSO protein predictors of metabolites. LASSO hits for each metabolite were mapped onto CORUM³⁵ and BioPlex³⁶. If a newfound LASSO protein predictor of a metabolite physically interacts with a protein known to regulate this metabolite via a known RHEA, TCDB, or Reactome network, then the LASSO hit was listed as potentially regulating the metabolite through the known network (Supplementary Table 5c). (o) Annotation of LASSO protein predictors of metabolites based on whether the proteins were known to be metabolic enzymes, transporters, and mitochondrial proteins. (p) Top 10 proteins in BAT and liver that predicted the highest number of metabolites. (OLS modeling for selected LASSO variables and two-sided t-test to calculate P values. P values adjusted by the Benjamini-Hochberg procedure in a, d, e, h-j, l).