Structure Supervised learning

Genetic programming, a specific form of evolutionary computing, has recently been used for predicting oral bioavailability [23], The results show a slight improvement compared with the ORMUCS Yoshida-Topliss approach. This supervised learning method and other described methods demonstrate that at least qualitative (binned) predictions of oral bioavailability seem tractable directly from the structure. 

Fritzke, B. (1994) Growing cell structures-a self-organizing network for unsupervised and supervised learning. Neural Networks 7 1441-1460... 

Back-propagation networks have been used in supervised learning mode for structure elucidation [41,42], A recall test with a separate data set confirms the quality of training. Novic and Zupan doubted the benefits of back-propagation networks for infrared spectroscopy and introduced the use of Kohonen and CPG networks for the analysis of spectra-structure correlations. 

Fig. 8.7 The Biomarker Patterns Software. This supervised learning program performs multivariate analyses and creates tree-like structured decision diagrams by splitting the original data set into sub-groups of highest possible purity. The program calculates which signals (i.e., proteins) are best suited to act as splitters, and the splitting rules define what intensity (i.e., protein abundance) a signal must have to be sorted into one...

There are two main classification systems to organize proteins based on their structure CATH [174] and SCOP [175]. These systems are used to label training data for a number of supervised learning problems found in protein structure prediction. This problem is divided into three subproblems depending on the data... 

The aim of QSPR/QSAR work is to develop mathematical models based on known cases to allow predictions for unknown cases. Furthermore, such models may lead to a better understanding of the often complex causal dependence of a property on structure. Important mathematical tools in the search for QSPRs are the statistical methods of supervised learning. Application of such methods requires a sufficiently large database of the appropriate structure-property pedrs. 

Initially, the structures of the real library are mapped onto real numbers using molecular descriptors. A vector of equal length of real numbers is obtained for each structure. Further, an experimental value of the property of interest is associated with each structure. Descriptor and property values are the input for statistical methods of supervised learning. These result in a predicting function, a function that is determined to lit the experimental property values reasonably weD (see Chapter 6). The predicting function requires a vector of descriptor values as input and returns a predicted property value as output. The complete process of descriptor calculation, the search for and hnally application of a predicting function is known as QSPR/QSAR research. 

Like most supervised learning methods, the goal of the decision tree methodology is to develop classification rules that determine the class of any object from the values of the object s attributes. In the case of decision trees, as the name implies, the classification rules are embodied in a knowledge representation fonnalism called a decision tree. This method has been used to derive structure-activity relationships and to learn classification rules for reactions. 

Two types of pattern recognition technique " have been applied to spectrum interpretation. Supervised methods (see Supervised Learning) are limited to one or more predefined structural classes and require representative training sets for each to develop the classifier. Unsupervised methods (see Unsupervised Learning) partition a set of spectra into clusters with common structural features on the basis of spectral features alone. No predefined classes are required. 

---