MetaboAnalyst contains 17 methods for curve fitting between individual metabolic features and continuous exposures (such as those in observational exposomics studies) based on the well-established drc R package. In general, we don’t know which model will be the best. MetaboAnalyst will evaluate for all selected models separately, and then use AIC to find the model with the lowest AIC.
These functions differ in their shapes and parameters.
Model data with Sigmoid (S-shaped) response
- L3/LL3/LL23: log-logistic function (three-parameter)
- L4/LL4/LL24: log-logistic function (four-parameter)
- L5/LL5/LL25: log-logistic function (5-parameters)
Model data with a decreasing response
- W13/W14/W23/W24: Weibull functions (3/4-parameters)
Model data with U-shaped response
- BC4 / BC5: Brain-Cousens (hormesis) (4/5 parameters)
Model data with an increasing response
- AR3: Asymptotic regression model (3 parameters).
- MM3: Michaelis-Menten model (3 parameters)
For more technical details, please refer to the drc R package.