engleski

Estimation of chance accuracy in classification structure-property models

For each model, including classification structure-property models related to small molecules or proteins, it is possible to calculate (or to estimate by simulations) the level of accuracy is obtained with randomly generated data or by a random model. However, in generation of random data one must ensure that random data have structure and distribution similar to those of real inputs. We present the analysis of several sets of data relating to the modeling/prediction of (1) the secondary structure of membrane proteins (based on their primary structure) and (2) relationship between the structure and properties of small molecules. All these problems have two or more classes, i.e. defined and experimentally determined groups. Obtained results show that the accuracy which is obtained by chance (with randomly generated data, and/or by a random model), is determined to a large extent by the distribution of experimental input data (classes) and by the (im)partiality of the model (the total numbers of individual classes predicted by impartial model are almost the same as those in experimental data).

classification model ; two-state model ; structure-property model ; structure prediction ; chance correlation ; membrane protein ; secondary structure

nije evidentirano