Title: Remarkable combinations of data types in the AmP collection
Authors: Kooijman, Augustine, Lika, Marn, Kearney
Affiliation: A-LIFE Amsterdam Institute for Life and Environment, Faculty of Science, VU University Amsterdam, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands
Date: 2023/03/21
Event: DEB2023

As is known, DEB parameters can only be estimated from combinations of several data types, rather than from a single data set.
This poses strong constrains on what function of data, predictions and weight coefficients can be minimized to arrive at parameter estimates: the loss function.
The classic regression loss function is of little use, due to differences in dimensions of the various data sets.
It is not a distance measure between data and predictions, since they cannot be interchanged without affecting the value of this loss function.
The symmetric bounded loss function is such a distance-measure:
the mean relative error for some 15 data sets per species and over 4000 species is only 0.05.
Many of the data set combinations in the AmP collection are simple, but some are remarkable in covering diverse aspects of metabolism,
  or showing how changes in food intake and temperature translate to changes in growth, reproduction or respiration.
This lecture discusses some of them briefly, to illustrate the challenges of estimating DEB parameters in practice and to show the predictive power of simple models.
Take, for instance, the entry for Alburnus alburnus, with the data sets dry weight as function of time and respiration and feeding rate as functions of dry weight.
One needs DEB theory to see why and how their combination determines the digestion and defecation efficiencies.
They turn out to be much lower and higher, respectively, than the default values.