Design, development and utilization of novel TAE descriptors and machine learning methods

Cyclazocine was used in early 1970's as an analgesic and a possible treatment for preventing relapse in post-addicts of heroin. However, further clinical researches were ceased because of its short duration of analgesic action. Cyclazocine analogues, especially 8-aminocyclazocine analogues were found able to retard this metabolic inactivation and increase its duration of action. From the most satisfactory result obtained from PEST descriptors with feature selection using Bootstrap-PLS, Q2 for training set is 0.06 and Q2 for testing set is 0.03, we can see that PEST descriptors reflect the electronic property as well as shape information complimentarily of the receptors and the binding site, and have promising application in QSAR modeling.; To better represent compounds in the models, improved descriptors are needed. I designed, developed and evaluated PEST Autocorrelation Descriptors (PADs) and PEST Chiral Descriptors (PCDs) on the basis of PEST shape descriptors. These two descriptors have shown enhanced ability for QSAR modeling. Also I designed, developed and evaluated TAE Type Fingerprint (TTFP) by comparing it with commercial fingerprints. TTFP is an electronic based fingerprint, which is calculated using TAE library and expressed in the sequence of positions of atom pairs. The major characteristic of TTFP, compared with other commercial fingerprints, is that it is based on atom pairs instead of the structural fragments. The prominent feature of TTFP is that it is good at predicting not only true positive activities but also true false activities.; UniSurCoMFA from Optive Inc. is evaluated and compared with CoMFA. 10 datasets, 5 charges and 2 alignment rules were used in the assessment. UniSurCoMFA appears a better 3D QSAR technique than CoMFA. The values of CV-R2 for general cases are higher when using UniSurCoMFA rather than CoMFA. Also, UniSurCoMFA performs a quicker and easier modeling calculations than CoMFA does because few parameters need to be optimized in UniSurCoMFA.