Extraction of Comprehensible Logical Rules from Neural Networks. Application of TREPAN in Bio and Chemoinformatics

摘要

TREPAN is an algorithm for the extraction of comprehensible rules from trained neural networks. The method has been applied successfully to biological sequence (bioinformatics) problems. It has now been extended to handle chemoinformatics (QSAR) datasets. The method has been shown to have advantages over traditional symbolic rule induction methods such as C5. Results obtained for bioinformatics and chemoinformatics problems using the TREPAN algorithm are presented.%TREPAN je algoritam za izlučivanje razumljivih pravila iz neuronskih mreža nakon provedenoga postupka učenja. Metoda je uspješno primjenjivana na probleme u bioinformatici, za analizu bioloških sekvencija. Primjena TREPAN metode sada se proširuje i na analizu skupova podataka u kemoinformatici (QSAR). Pokazano je da metoda ima prednosti u odnosu na uobičajene postupke koji se rabe za indukciju simboličkih pravila poput metode C5. Prikazani su rezultati koji su dobiveni u analizi bioinformatičkih i kemoinformatičkih problema s pomoću algoritma TREPAN.