一种基于进化学习中药定量组效关系建模方法.pdfVIP

39 4 ( ) Vol. 39 N o. 4 2005 4 Journ al of Zh ej an g Un vers ty( En g n eer n g Sc en ce) Apr . 2005 刘雪松, 王 毅, 程翼宇 ( , 3 10027) : , ( QCAR ) . , ( BP) , , ( GA ) , QCAR . . , , , . : ; ; ; ; : O657 . 7 ; TP 18 : A : 1008 973X( 2005) 04 04 95 05 Approach to bioactivity prediction of hebal medicine using evolutional learning2based quantitative composition2activity relationship model LIU Xue2song, WANG Y , CH ENG Y 2yu (Department of ChineseMedicine Science and Engineering, Zhej iang University, Hangzhou 310027, China) Abstract: An ev olut onal learn n g m et h od was proposed t o est abl sh a quant t at ve comp os t on2act v t y relat onsh p ( QCAR ) model an d pred ct t he b oact v ty of Ch n ese m ed c ne. Int egrated w t h back prop agat on ( BP ) algor t hm, a novel adapt ve float cod n g gen et c algo thm ( AF CGA ) was used t o t ra n n eural net work n QCAR model n g. T h e present ed m et h od mproves t he capab l t y of obt a n n g th e glob al opt mum n n eural computat on an d el m n ates th e t runcat on error produ ced n th e process of encod n g an d decod n g for class c gen et c algor t hms( GA ) . H en ce, th e t ra n n g an d pred ct n g accuracy of t he OCAR m odel was mproved great ly. A s an ex amp le, st udy of Dang ui ver f ed t he pred ct ng ab l lt y of t h e prop osed m et h od . T he r esult s showed t hat , n o mat ter n Cross Val dat on or In dep en dent Test, t h e p er for mance of t h e present ed model s gn f cant ly excels th at of ot h er convent on al m odels. A nd t h s approach can b e u sed to pred ct b oact v t y of n at ural dru gs from ch em ca