Figure 8

The covariate effects inferred by Model 1 (All peaks) were more robust to noise compared to Model 3 (Single peak). At moderate levels of noise, which is the regime of many biological experiments, the confidence intervals over 100 randomizations did not overlap at all. The robustness was quantified as the Spearman correlation (y-axis) between the effects inferred from the noisy and noise-free versions of the lipidomic gene silencing data set as a function of the level of noise (x-axis).