Debber - Frequently Asked Questions
How do I use Debber results?
Try to avoid using only a "best estimate" such as the median. The estimates are subject to considerable uncertainty, which can be seen from the percentiles
shown when you click a parameter and from the spread in model results in ensemble analyses.
Instead, we recommend you use all information - best estimates and their uncertainty - ideally in combination with new data specific to your species.
For instance, you can use the provided information to set plausible ranges (e.g., the 2.5th - 97.5th percentile) for the primary DEB parameters during calibration to new data,
or use the probability distribution (multivariate normal; mean and covariance matrix downloadable underneath the table) as prior in a Bayesian analysis.
Such a Bayesian analysis is performed automatically if you prescribe the value of life history traits as part of your query.
If you do not have additional data
on your species, we suggest you use an ensemble of models, constructed by drawing from the parameter probability distribution.
Are Debber results reliable? (how are they quality-controlled?)
Debber performs two tasks: first, it estimates DEB primary parameters through phylogenetic analysis of the "Add-my-Pet" database.
Second, it analyzes DEB models to compute traits and simulate growth, reproduction and survival. Both stages are quality controlled.
We verify whether Debber's parameter estimates are reliable through cross-validation. This compares the predicted
distribution of primary parameters against true (Add my Pet) values. For instance, it verifies whether DEB parameters for a new species
indeed fall 95 out of every 100 times within Debber's 95% confidence interval.
We check the logic for trait prediction and model simulation ("pydeb") by comparing predicted traits against the trait values provided by "Add-my-Pet".
Are Debber results reproducible? (will they be the same when I rerun my query, even after several months?)
Debber's prior distribution of DEB parameters is a deterministic result and therefore fully reproducible.
This includes all results shown under "DEB parameters" if you do not provide additional information (life history statistics)
as part of your query. The other results - trait distributions, time series of growth, reproduction and survival - are produced using
stochastic ensemble methods (Monte Carlo sampling when you provide the taxon name only; Markov Chain Monte Carlo sampling when
you provide life history statistics as part of your query). These will differ slightly upon every query.
Additionally, the datasets upon which Debber is built (notably, snapshots of the Add-my-Pet collection) are regularly updated. This
will change Debber results, though as long as Add-my-Pet only adds new entries, changes can be expected to be small and lead to improved
results. As we value reproducibility, it is always possible to revert to a previous value of the Debber database to exacty reproduce
(the deterministic parts of) a past result. To make this possible, Debber provides a permalink
to your result at the top of the page; this link includes the specific version of the Debber database and can be used to reproduce your result even
after Debber updates.
What is the accuracy of Debber results?
As Debber performs tens to hundreds of thousands of simulations on the fly, it has to carefully balance computational cost and accuracy.
All numerical algorithms are currently configured to produce results with a maximum error of 1%. We believe this is sufficient for most biological applications.
Why do Debber estimates differ from the values for the same species in the Add-my-Pet collection?
When analyzing the DEB parameters across all taxa, we do not only expect phylogenetic varation (parameters of closely related species are similar,
those of distantly related species diverge), but also phenotypic variation: "observed" parameter values
can deviate from the mean of the species due to intraspecific variation, sampling error,
and parameter estimation uncertainty. Thus, AmP parameters are assumed uncertain (and this uncertainty is quantified).
As a result, Debber's best estimate for a species will not only reflect its AmP values, but also circumstancial information
- in particular, AmP values of closely related taxa.
Which typified models are supported?
Typified models use all parameters of the standard model, which retain their original interpretation, and add additional parameters to describe
specific phenomena (e.g., the onset of particular behaviour as a maturation or time threshold). Therefore, the estimates of the standard DEB parameters
provided here can be used for typified models too. If parameters specific to your typified model of choice are not estimated themselves,
you will need to fill these in on your own. Debber allows you to pick a typified model upfront. By default, it
will use the "abj" typified model, unless the selected taxon is a non-egg-laying mammal (that is, the taxon belongs to any order from
class Mammalia, except Monotremata). For non-egg-laying mammals, Debber will account for fetal development by using the "stf" typified model.