Pattern recognition with QSAR

There is a tendency to think of medicinal chemistry as primarily a logical exercise. A specific and trivial example would be the much maligned QSAR exploration of methyl, ethyl, butyl, futile. This author believes that equating medicinal chemistry with QSAR is incorrect. There is a definite place for what might for want of a better term be called high-order pattern recognition. A specific example is the time tested... 

On the other hand, several investigators (6, 7) have taken another approach, based on pattern recognition. These dichotomous models search for agreement between dependent variables i.e., whether a chemical entity or substructure can be associated with a particular toxic property. For example, certain N-nitrosamine groups are associated with tumors in animals. Since this consideration is not dependent on a relationship between the endpoint and the dose, the quantitative term is dropped from QSAR and the effort simply named SAR. This approach is best expressed by the dependent equation ... 

Various attempts have been made to use pattern recognition [24, 25] in QSAR studies and successful applications have been reported. Soft modeling techniques, e.g. the partial least squares (PLS) method [26, 27], now offer better opportunities. With the help of this principal component-like method the explanatory power of many, even hundreds or thousands of variables can be used for a limited number of objects, a task being absolutely impossible in regression analysis in which the number of objects must always be larger than the number of variables. 

Experience with the pattern recognition technique points out that even the most elaborate QSAR study backed by all sorts of data manipulation and statistical analysis can bear little fruit if the molecular properties do not have some rational, causal connection with the biological effect under consideration. In seeking correlations with biological data the molecular features must have some relevance to the biochemistry involved in the drug eliciting its biological response. 

In SAR research, the purpose is to connect the biological activities of a series of compounds with their physico-chemical properties by using regression analysis, pattern recognition methods, or other sophisticated data processing methods. Generally speaking, the activities and properties are connected by a function F as follows for QSAR problems ... 

---