Jurs Research Group
ADAPT is a software system for the development of quantitative structure-activity relationships (QSARs) or quantitative structure-property relationships (QSPRs). It implements an inductive approach where the QSPRs or QSARs are developed from a set of known values for compounds forming a training set. The QSPRs or QSARs are generated in a three-step process:
- representing the molecular structures by calculated descriptors
- selecting the best subset of descriptors using feature selection methods
- mapping these descriptors onto the known biological activity or physical property using regression analysis or computational neural networks.
The studies involve having a training set of at least 30 to 50 compounds with known values, and larger training sets are preferable.
The creation of QSARs and QSPRs involves the following individual steps:
- graphical entry and storage of molecular structures and their associated data
- generation of three-dimensional molecular models
- molecular structure descriptor calculation
- feature selection
- analysis of the descriptors using multivariate statistical, pattern recognition, or neural network methods to build predictive equations.
ADAPT supports all of these activities. In addition, it can import structures as molfiles and import 3-D coordinates as well.
ADAPT has a large selection of molecular structure descriptor generation routines. The four general classes of structural descriptors are topological, geometical, electronic, and physicochemical representations of the molecules. Several hybrid descriptors that combine features of these classes are also available.
Analysis of the descriptors can be done by several different approaches depending on the characteristics of the data being analyzed. When quantitative property or activity data are available, the descriptors can be related to the desired property or biological activity by multiple linear regression analysis which yields quantitative predictive models. Computational neural networks are used to develop quantitative non-linear models that link the structural descriptors to the activity or property of interest. When qualitative property or activity data are being analyzed, then the descriptors can be related to the property or biological activity by pattern recognition methods in order to develop discriminants or clustering definitions. By whichever method they are developed, the models then can be used for predictive purposes and for gaining insight into which molecular features are most important in determining the property or activity values.
ADAPT is written in Fortran and is installed on a DEC alpha workstation under the Unix operating system. It has also been ported to Linux using the Intel Fortran compiler. A number of scripts to automate various tasks are also available. Approximately 120 MB of storage are required for source code, libraries, executables, and documentation. More information regarding ADAPT can be found here (choose the ADAPT option from the menu on the left)