| Summary: | In the development of drugs compounds suitable for human being, many experiments have to be conducted to ensure drugs safe consumption and generally takes almost 10 to 12 years for a particular drugs to enter the market from laboratory. Therefore, the pattern recognition in QSAR is significant for analyzing the data and developing several necessary models, so that only novel drugs candidate will be synthesized. There are three important aspects for the classification of BBB activity in this work, (1) variable reduction by PCA (2) variable selection and class separation with comparison of three methods such as T-Statistics, Partial Least Squares Regression Coefficient (PLSRC) and newly invented Self Organising Maps Discriminatory Index (SOMDI). and (3) classification, a comparison of linear (PLSDA) and non linear (SuSOMs) methods. The number of PCA component determined by LOO cross-validations is seven. Based on PCA score, the variables selected by T-Statistics and SOMDI are more selective and can provide better separation for BBB activity than PLSRC. Models performances and validations, built through PLSDA and SOMs show that the consensually selected 7 descriptors in this work by using SOMDI, T-statistics and PLSRC were able to classify BBB penetration and non-penetration compounds. © 2015 IEEE.
|
|---|