26 KB148 KB17 KB199820252010Héberger, KárolyMinovski, NikolaVračko, Marjanštevilka:3letnik:57str. 586-590ISSN:1318-0207COBISSID:4510490URN:URN:NBN:SI:doc-K2S5PLJPenSlovenian Chemical SocietySlovensko kemijsko društvoActa chimica slovenicafluorokinolikemijska strukturamodeliranjenevronske mrežeQSARQuantitative Structure–Activity Relationshipzdravilne učinkovineRanking of QSAR models to predict minimal inhibitory concentrations toward Mycobacterium tuberculosis for a set of fluoroquinolones|CP-ANN technique was used to build 54 different QSAR models. The models were built for three sets (assays) of fluoroquinolones considering their antituberculosis activity and using different technical parameters (dimension of network and number of learning epochs). The models served as a reliable basis for ranking by a new powerful method based on sum of ranking differences (SRD). With the applied SRD procedure we can find the optimal ones. The best model can be selected easily for the first assay. Two models can be recommended for the second assay, and no recommended model was found for the assay3V prispevku smo analizirali 54 QSAR modelov, ki so bili narejeni s protitočnimi nevronskimi mrežami (CP-ANN). Modeli so bili narejeni na treh nizih fluorokinolonov kot potencialnih antituberkuloznih učinkovinah. Modele smo razvili ob upoštevanju različnih tehničnih parametrov, kot sta dimenzija nevronske mreže in število učnih epoh. Na nizu modelov smo uporabili pred kratkim predlagano metodo za hierarhično urejanje modelov (SRD), s katero lahko poiščemo najbolj zanesljive modele. Z metodo SRD smo za prvi niz izbrali en model, za drugi niz smo predlagali dva modela, za tretji niz pa nismo našli modela, ki bi bil boljši kot ostali modeliTEXTznanstveno časopisjejournalsSlovenian National E-content AggregatorNational and University Library of SloveniaSlovensko kemijsko društvo