=20
The key abstract concept used in this work for assessment of AD is dista=
nce to model (DM), defined as follows:
Distance to a model is any numerical measure of the prediction u=
ncertainty for a given compound by the model.
A distance to model assesses how =E2=80=9Cfar=E2=80=9D is the compound f=
rom the model. The compounds that are =E2=80=9Cfurther from the model=E2=80=
=9D, which have larger values of DM, are by definition expected to have low=
er prediction accuracy than compounds that have smaller values of DM. It sh=
ould be clearly stated that prediction accuracy correlates
with DM only in average: for example, compounds with DM in range [0.5, 0=
.6] will on average have higher prediction accuracy than compounds with DM =
in range [0.6, 0.7] but, nonetheless, the prediction errors for some compou=
nds from the first interval can be bigger than for some compounds from the =
second interval. In other words, the key property of a DM is the discrimina=
ting ability, i.e. the ability to discriminate predictions of high and low =
accuracy. Importantly, DMs estimate the reliability of predictions. While a=
ccuracy is an objective measure that has a rigid calculation procedure, rel=
iability is subjective and can be estimated in numerous ways. Therefore, th=
ere is a number of different DMs that assess the reliability of predictions=
from different perspectives. Here, we briefly overview the DMs used for th=
e AD assessment in this work.
http://mediatum.ub.tum.de/node?id=3D1004002&nb=
sp;
=20