Chemistry Reference
In-Depth Information
In the second (2) or pharmacophore method, active compounds are identified by
matching molecules to an assembly of steric and physicochemical features for
activity. This method complements docking by its ability to select active
compounds of higher structural flexibility but is less effective in modeling
detailed binding interactions. It is also sensitive to training datasets, quality of
conformational sampling, multiple choices of features, molecular overlay, binding
affinity estimation and anchoring point selection.
The third (3) QSAR method makes usage of statistically significant correlation
between molecular structures and activities to identify active compounds. Specific
sets of structural and physicochemical properties (descriptors relevant to
activities) can be used to represent molecular structures. This model is dependent
on representativeness of structure-activity data, concept compatibility, and
influence of data outliers, starting geometry, quantitative relationships and
multiple choices of solutions.
The fourth (4) or similarity searching technique measures the level of structural
similarity to known compounds in order to identify novel active compounds. The
technique uses methods such as molecular fingerprints, molecular descriptors and
molecular structural similarity. The technique is effective and fast but limited
however to requirements of structural or sub-structural similarity to known active
compounds.
The fifth (5) method is a machine learning method, which includes binary kernel
discrimination and support vector machines using statistical analysis of intrinsic
correlations between activities and the structural/physicochemical profiles of
known compounds. The objective is to identify the active from the inactive
compounds. The technique uses statistical models to predict diverse spectrum of
structural and physicochemical properties at high CPU speed useful for screening
large libraries. The method depends on training set diversity and parameter
ranges.
High demands on screening speed and low false hit rates are placed on virtual
screening due to significantly large libraries involved. Consequently, VS methods
are used in combination with each other and with constraints/filters. These include
