Fig. 1

Noise in experimentally determined biomass severely distorts gLVM parameter estimation. a Scatter plot with fitted linear regression line for 2 16S qPCR technical replicates from Bucci et al. b Copy number variation for 16S rRNA genes in members of 4 major phyla of human gut bacteria. c Relative impact of different experimental (qPCR_rep1, 1 qPCR technical replicate; qPCR_rep3, mean of 3 qPCR technical replicates) and computational (RA, relative abundance; CSS, CSS normalization) data scaling approaches on gLVM parameter estimation (BVS algorithm for MDSINE), in comparison with using noise-free biomass or using BEEM. Boxplots represent the summary of 15 simulations (10 species, 30 replicates with 30 time points each), and 3 different metrics are shown here including median relative error for growth rate (μ) and interaction (β) parameters, and AUC-ROC for the interaction network. Dashed horizontal lines represent the performance of randomly generated parameters from the simulation model