Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Points represent individual samples colored by pool size and shaped by age. PCA was performed on mean-centered metabolite values without additional scaling. Variance explained by each PC is shown in parentheses on axis labels.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques:

Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Mean distance (± SE) between all pool-size pairs (5–100, 5–50, 50–100) is shown for each metabolite panel, coloured by strain. Individual replicate values are shown as transparent points.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques:

Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Points represent individual samples colored by pool size and shaped by diet. PCA was performed on mean-centered metabolite values without additional scaling. Variance explained by each PC is shown in parentheses on axis labels.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques:

Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Pairwise Euclidean distances between pool sizes in the full metabolomic space. Mean distance (± SE) between all pool-size pairs (5–100, 5–50, 50–100) is shown for each metabolite panel, colored by diet.Individual replicate values are shown as transparent points.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques: Metabolomic

Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Retention of true and false positive metabolite detection across replicate and pool-size downsampling. The proportion of diet-associated metabolites identified under downsampling was expressed as a percentage of the reference set defined at PoolSize = 100 with full replicates (8 per diet; 100%). True positives (solid lines) represent metabolites that were significant (FDR < 0.05) in both the downsampled and reference datasets, while false positives (dashed lines) were significant only in the downsampled condition. Lines show the mean percentage across all combinations of replicate removal, and shaded ribbons indicate the interquartile range (25th–75th percentile), reflecting variability across downsampling iterations. Across all metabolite panels, reductions in replicate number and pool size led to a progressive loss of true positives, while false positives remained comparatively low, indicating reduced statistical power rather than systematic inflation of spurious detections under downsampling.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques:

Journal: bioRxiv

Article Title: Signal, noise, and sampling: How pool size and replication shape metabolomic inference

doi: 10.64898/2026.04.07.717001

Figure Lengend Snippet: Metabolites were grouped into high, medium, and low effect-size bins based on tertiles of absolute diet effect sizes estimated from the full dataset (PoolSize = 100, full replicates), calculated separately within each metabolite panel. For each bin, the proportion of metabolites remaining significant (FDR < 0.05) was evaluated across all combinations of replicate downsampling and pool sizes (5, 50, and 100). Lines represent the mean fraction of metabolites remaining significant across all downsampling iterations, and shaded ribbons indicate the interquartile range (25th–75th percentile), reflecting sensitivity to which replicate populations were removed. Across all panels, metabolites with larger effect sizes exhibited greater robustness to reductions in replicate number and pool size, whereas low-effect metabolites rapidly lost significance under downsampling.

Article Snippet: Metabolites were identified based on accurate mass, MS/MS spectra, isotope pattern, and retention time using PeakView and MultiQuant software (AB Sciex) compared with metabolites in the facility’s metabolite IROA library (650 standards).

Techniques: