Chemistry Reference
In-Depth Information
1.
1D descriptors (depend on the formula of the molecule and can only
give information on the composition of the element or molecular
weight);
2.
2D descriptors (obtained from the connectivity graph or a molecular
graph);
3.
3D (include three-dimensional geometric information of a molecule).
Currently, only one 1D descriptor, the molecular weight (MW), is useful during
absorption and bioavailability prediction. Regarding 2D descriptors there are
several options, as they are quickly calculated. 2D descriptors include topological
polar surface area (TPSA), number of hydrogen bond acceptors (NHBA), number
of hydrogen bond donors (NHBD), number of hydrogen bond donors and
acceptors (NHD), octanol-water partitioning coefficient (logP), apparent partition
coefficient (logD), intrinsic solubility (logS), number of rotatable bonds (Nrot),
number of molecular fragments, electrotopological state index (E-state), and a
variety of other topological parameters. Lastly, the 3D molecular descriptors most
widely used include Polar Surface Area (PSA), molecular surface area (MSA) and
molecular volume (MV) [9].
Hou
et al
. [4] studied the performance of a support vector machine (SVM) to
classify compounds with high or low fractional absorption (%FA > 30% or %FA
≤ 30%). For this, 10 models of SVM classification were considered to investigate
the impact of different individual molecular properties on %FA. Among them
were the topological polar surface area (TPSA), octanol-water patitioning
coefficient (logP), apparent partition coefficient at pH = 6.5 (logD6.5), number of
violations of the Rule of Five (Nrule-of-five), number of hydrogen bond donors
(NHBD), number of hydrogen bond acceptors (NHBA), intrinsic solubility (logS),
number of rotatable bonds (Nrot), molecular volume (MV), and molecular weight
(MW). The database used for analysis consisted of 648 chemical compounds of
which 579 molecules were believed to be transported by passive diffusion.
First, the 10 classification models were built using each descriptor individually.
The RBF kernel function was used in the analysis of SVM. Subsequently, a
