Amplified descriptor

Structure and function need to be jointly considered in the assessment of effects of stressors on river systems. It has been shown that the two sets of parameters offer complementary information since they cover different time scales and responses. This being shown in the case of biofilms is not a unique characteristic of them, but it might be applied to all other biological communities (e.g. macroinvertebrates, fish). These differ from the biofilm in its higher size and life span, and therefore in their integrative capacity to reflect effects in one part of the ecosystem. Higher traffic levels in addition to biofilms should be considered to study the whole ecosystem. In all of these biological compartments, the combined use of descriptors may amplify our ability to predict the effect of stressors on river basins. 

As the previous section illustrates, descriptor selection is an important concept. The problem is amplified due to inherently high-dimensional representations (dne to large numbers of descriptors), which have a significant impact on the speed of computation... 

Conseqnently, the resulting RDF descriptors — here called the amplified and attennated RDF — exhibit quite different shapes. The attenuated RDF descriptors typically exhibit more, or significant, peaks than the amplified variant when dynamic properties are used. In particular, dynamic properties of carbon atoms are often small in comparison with one of the heteroatoms. Thus, peaks related to carbon atoms are also small, whereas the peaks related to the heteroatoms are intensified. 

In general, the property product leads to a descriptor that emphasizes extreme differences in properties — thus the name amplified. In most of the investigations presented in this volume, the product of properties was used to gain advantage of this effect for the specific task. 

Amplified RDF is a descriptor resulting from using a property product as a factor for the exponential term in an RDF descriptor, leading to an amplification of peaks that originate from atom pairs with strongly different atom properties. 

Each X vector may also be scaled with a suitable factor to take into account for example different units for the variables. This, however, is non-trivial and requires careful consideration. A common procedure, which avoids a user decision, is to normalize each X vector to have a variance of 1, a procedure called autoscaling. Antoscsil-ing equalizes the variance of each descriptor and can thus amplify random noise in the sample data and reduce the importance of a variable having a large response and a good correlation with the y data. 

FID. See Flame-ionizatiOTi detection (FID) Field-amplified sample stacking (FASS), 1189 Fine emery, dry and grippy (astringency descriptors), 2266 Fine/flavor cocoas, 1601 Fine fragrances, 3291 Fingerprinting, 1022 Firmicutes, 2441, 2479 First-pass metabolism, 2322 Fish, 4069, 4079... 

---