PRACA ORYGINALNA · SZTUCZNE SIECI NEURONOWE 665 Tom 79 · nr 11 · 2023 Application of Artificial Neural Networks for Predicting Imidazole Derivatives’ Antimicrobial Activity against Enterococcus faecalis Anna Badura 1 , Łukasz Pałkowski 2 , Alicja Nowaczyk 3 , Marta Poćwiardowska-Głąb 4 , Adam Buciński 1 1 Katedra Biofarmacji, Wydział Farmaceutyczny, Uniwersytet Mikołaja Kopernika w Toruniu, Collegium Medicum im. Ludwika Rydygiera w Bydgoszczy, Polska 2 Katedra Technologii Postaci Leku, Wydział Farmaceutyczny Uniwersytet Mikołaja Kopernika w Toruniu, Collegium Medicum im. Ludwika Rydygiera w Bydgoszczy, Polska 3 Katedra Chemii Organicznej, Wydział Farmaceutyczny, Uniwersytet Mikołaja Kopernika w Toruniu, Collegium Medicum im. Ludwika Rydygiera w Bydgoszczy, Polska 4 Zakład Diagnostyki Laboratoryjnej, Wojewódzki Szpital Dziecięcy im. J. Brudzińskiego w Bydgoszczy, Polska Farmacja Polska, ISSN 0014-8261 (print); ISSN 2544-8552 (on-line) Application of Artificial Neural Networks for Predicting Imidazole Derivatives’ Antimicrobial Activity against Enterococcus faecalis Artificial neural networks (ANNs) have emerged as a valuable tool in facilitating the design of synthesis and guiding subsequent biological experiments in the systematic exploration for novel antimicrobial agents. In this paper, two multilayer perceptron-type neural networks (MLP) are designed to predict the biological activity of compounds based on their physicochemical properties and structure. This approach was tested against Enterococcus faecalis using a series of 140 imidazole derivatives. The use of quaternary ammonium salts in this research originated from their acknowledged ability to act as antiseptics and disinfectants. Additionally, they were considered promising in addressing various microorganisms, including Gram-positive bacteria. The designed regression model accurately predicted the minimum inhibitory concentration for E. faecalis growth. The coefficient of correlation between the actual values and the network predictions for the training set was R = 0.91, for the test set was R = 0.91, and for the validation set was R = 0.97.Additionally, the classification model successfully categorized the tested compounds as predictively active or inactive against the targeted microorganism (classification accuracy: 92.86%). Sensitivity analyses highlighted specific molecular descriptors Corresponding author Anna Badura, Katedra Biofarmacji, Uniwersytet Mikołaja Kopernika w Toruniu, Collegium Medicum im. Ludwika Rydygiera w Bydgoszczy, Wydział Farmaceutyczny, ul. Jurasza 2, 85-089 Bydgoszcz, Polska; e-mail: anna.badura@cm.umk.pl Sources of financing No sources of financing were indicated. Conflict of interest No conflicts of interest. Received: 2023.11.30 Accepted: 2024.01.26 Published on-line: 2024.02.07 DOI 10.32383/farmpol/183146 ORCID Anna Badura – 0000-0002-6560-8231 Łukasz Pałkowski – 0000-0002-8219-6339 Alicja Nowaczyk – 0000-0003-4945-2369 Marta Poćwiardowska-Głąb – 0009-0008-9920-5116 Adam Buciński – 0000-0002-0558-9139 Copyright © Polish Pharmaceutical Society This is an open-access article under the CC BY NC license https://creativecommons.org/licenses/by-nc/4.0/