NTP Research Report on National Toxicology Program Approach to Genomic Dose-Response Modeling

ALOHA: Aggregated local extrema splines for high-throughput dose–response analysis

“…Smaller BMDU/BMDL estimates indicate increased certainty in BMD estimates and more reliable fits. We define probes with reliable expression activity as those having a (i) BMD less than the maximum dose and (ii) 95% confidence limit BMDU/BMDL less than 40, which are consistent with the reliability criteria proposed by NTP et al [9]. When fitting parametric models, the best model fits must also have a goodness-of-fit p-value > 0.1, which indicates no significant difference between the observed expression values and those estimated by the model.…”

“…In this definition, the BMR indicates a standard deviation change in gene expression from the control. For our application, we consider set BM R = 1, which is consistent with previous analyses with these datasets Ramaiahgari et al [33] and NTP et al [9].…”

“…To date, several GDR modeling approaches have been developed with a variety of methodological features, see S1 Table 1. Except for BMDExpress [6][7][8], which has documented guidelines in NTP et al [9], none of the other approaches have established an approach for determining a genomic POD linked to gene-and pathway-level biological alterations. Consequently, BMDExpress [6][7][8] is still the most widely utilized approach for GDR modeling and will be the focus of our comparison.…”

“…Consequently, BMDExpress [6][7][8] is still the most widely utilized approach for GDR modeling and will be the focus of our comparison. The platform uses a multi-step framework of parametric modeling and bioinformatics tools to characterize DR patterns and evaluate alterations in cellular function [6][7][8][9]. It addresses common challenges by using genomic data in risk assessment, but some of the assumptions made may not be appropriate given the observed data.…”

“…In cases where cytotoxicity occurs in high-doses, a non-monotonic (i.e., curve with umbrella orderings) is possible [10]. For such shapes, parametric models are typically inadequate, because most of the models assume monotone changes in expression [9,[11][12][13], which may lead to removing genes with biological relevance due to poor model fit. Another limitation of BMDExpress is the enrichment analysis used to detect altered cellular functions and obtain genomic PODs.…”