Empirical methods for determining the favorability of a
solvation site do so by analogy with known sites. A site is
evaluated and compared to a database of known solvation and
non-solvation sites, and predicted as being more similar to
one than the other. A set of features to observe and compare
between the solvation sites and non-solvated sites must be
selected prior to classification. We use our novel classifier
to distinguish the features of the data set that are the most
statistically relevant and to weigh these features appropriately
to aid in the classification of possible water coordination sites.
The paper is organized as follows. In Section II we present
an overview of the main results of the paper and give a
brief review of the fundamental concepts and related work.
Section III introduces the Bayesian discriminant function with
non-linear weighting and details a Gaussian smoothing factor
introduced to mitigate biasing sampling anomalies. Section
IV details our EC-based approach, while Section V reports
the experimental results of this approach on several medical
datasets and on the prediction of water solvation sites for
ligand docking. We conclude with a discussion of these results
in Section VI