Jerzy Krysiński

Artificial neural networks in prediction of antifungal activity of a series of pyridine derivatives against Candida albicans.

Quantitative structure-activity relationships (QSAR) studies of antifungal activity against Candida albicans of a large series of new pyridine derivatives were conducted with the use of artificial neural networks (ANNs). The application of ANNs has been provided with respect to the prediction of antimicrobial potency of new pyridine derivatives based on their structural descriptors generated by calculation chemistry. Antifungal activity against C. albicans has been related to a number of physicochemical and structural parameters of the pyridine derivatives investigated. The activity was expressed as logarithm of the reciprocal of the minimal inhibitory concentrations, log 1/MIC. Molecular descriptors of agents were obtained from structure fragment reference databases and by quantum-chemical calculations combined with molecular modeling. A high correlation resulted between the ANN predicted antifungal activity, log 1/MIC(pred), and that one from biological experiments, log 1/MIC(exp), for the data used in the testing set of pyridine was obtained with correlation coefficient, R, on the level of 0.9112.

Rough Set Theory and Decision Rules in Data Analysis of Breast Cancer Patients

In this paper an approach based on the rough set theory and induction of decision rules is applied to analyse relationships between condition attributes describing breast cancer patients and their treatment results. The data set contains 228 breast cancer patients described by 16 attributes and is divided into two classes: the 1st class – patients who had not experienced cancer recurrence; the 2nd class – patients who had cancer recurrence. In the first phase of the analysis, the rough sets based approach is applied to determine attribute importance for the patients’ classification. The set of selected attributes, which ensured high quality of the classification, was obtained. Then, the decision rules were generated by means of the algorithm inducting the minimal cover of the learning examples. The usefulness of these rules for predicting therapy results was evaluated by means of the cross-validation technique. Moreover, the syntax of selected rules was interpreted by physicians. Proceeding in this way, they formulated some indications, which may be helpful in making decisions referring to the treatment of breast cancer patients. To sum up, this paper presents a case study of applying rough sets theory to analyse medical data.

A data mining approach to optimize pellets manufacturing process based on a decision tree algorithm.

The present study is focused on the thorough analysis of cause-effect relationships between pellet formulation characteristics (pellet composition as well as process parameters) and the selected quality attribute of the final product. The shape using the aspect ratio value expressed the quality of pellets. A data matrix for chemometric analysis consisted of 224 pellet formulations performed by means of eight different active pharmaceutical ingredients and several various excipients, using different extrusion/spheronization process conditions. The data set contained 14 input variables (both formulation and process variables) and one output variable (pellet aspect ratio). A tree regression algorithm consistent with the Quality by Design concept was applied to obtain deeper understanding and knowledge of formulation and process parameters affecting the final pellet sphericity. The clear interpretable set of decision rules were generated. The spehronization speed, spheronization time, number of holes and water content of extrudate have been recognized as the key factors influencing pellet aspect ratio. The most spherical pellets were achieved by using a large number of holes during extrusion, a high spheronizer speed and longer time of spheronization. The described data mining approach enhances knowledge about pelletization process and simultaneously facilitates searching for the optimal process conditions which are necessary to achieve ideal spherical pellets, resulting in good flow characteristics. This data mining approach can be taken into consideration by industrial formulation scientists to support rational decision making in the field of pellets technology.

Antimicrobial Activity and SAR Study of New Gemini Imidazolium-Based Chlorides

A series of 70 new 3,3′(α,ω‐dioxaalkyl)bis(1‐alkylimidazolium) chlorides were synthesized. They were characterized with respect to surface active properties and antimicrobial activity against the following pathogens: Staphylococcus aureus, Enterococcus faecalis, Acinetobacter baumannii, Escherichia coli, Klebsiella pneumoniae, Enterobacter cloacae, Pseudomonas aeruginosa, Candida krusei, and Candida albicans. In this article, besides description of the synthesis, we characterize a set of features of these compounds, concerning their structure (described by the length of the dioxaalkan spacer and the length of the alkyl substituent in the aromatic ring) and surface active properties (critical micelle concentration, value of surface tension at critical micelle concentration, value of surface excess, molecular area of a single particle, and free energy of adsorption of molecule). Then, we present a SAR study for Staphylococcus aureus, as one of the most widespread pathogenic strains, conducted with the help of the Dominance‐based Rough Set Approach (DRSA), that involves identification of relevant features and relevant combinations of features being in strong relationship with a high antimicrobial activity of the compounds. The SAR study shows, moreover, that the antimicrobial activity is dependent on the type of substituents and their position at the chloride moiety, as well as on the surface active properties of the compounds.

Clinical data analysis using artificial neural networks (ANN) and principal component analysis (PCA) of patients with breast cancer after mastectomy

Summary Background Exploitation of the several types of information on patient, disease and treatment variables ranging from sociological to genetic ones by means of chemometric analysis was considered and evaluated. Aim Performance of modern data processing methods, namely principal component analysis (PCA) and artificial neural network (ANN) analysis, is demonstrated for predictions of the recurrence of breast cancer in patients treated previously with mastectomy. Materials/Methods The data on 718 patients were retrospectively evaluated. 11 subject and treatment variables were determined for each patient. A matrix of 718×11 data points was subjected to PCA and ANN processing. The properly trained ANN was used to predict the patients with recurrence and without recurrence within a 10-year period after mastectomy. Results It was found that the prognostic potency of the trained and validated ANN was reasonably high. Additionally, using the principal component analysis (PCA) method two principal components, PC1 and PC2, were extracted from the input data. They accounted cumulatively for 37.5% of the variance of the data analyzed. An apparent clustering of the variables and patients was observed – these have been interpreted in terms of their similarities and dissimilarities. Conclusions It has been concluded that ANN analysis offers a promising implementation to established methods of statistical analysis of multivariable data on cancer patients. On the other hand, PCA has been recommended as an alternative to classical regression analysis of multivariable clinical data. By means of ANN and PCA practically useful systematic information may be extracted from large sets of data, which can be of value for prognosis and appropriate adjustment of the treatment of breast cancer.

Artificial neural networks for prediction of antibacterial activity in series of imidazole derivatives.

Artificial neural networks (ANNs) have been applied for the quantitative structure-activity relationships (QSAR) studies of antibacterial activity against Escherichia coli, Serratia marcescens, Proteus vulgaris, Klebsiella pneumoniae and Pseudomonas aeruginosa of a large series of new imidazole derivatives. Antibacterial activity against individual bacteria, expressed as logarithm of reciprocal of the minimal inhibitory concentrations, log 1/MIC, has been related to a number of physicochemical and structural parameters of the imidazole derivatives investigated. Molecular descriptors of agents were obtained by quantum-chemical calculations combined with molecular modelling and from respective structure fragment reference data (e.g., log P). A high correlation resulted between the predicted from ANN model antibacterial activity, log 1/MIC(ANN), and that from biological experiments, log 1/MIC(exp), both for the data used in learning and in the testing sets of imidazoles. Correlation coefficient, R, depending on the type of bacteria and structural subset of analysed imidazole compounds, varies from 0.875 to 0.969. The applicability of ANNs has been demonstrated for the prediction of pharmacological potency of new imidazole derivatives based on their structural descriptors generated exclusively by calculation chemistry.

Simultaneous determination of ciprofloxacin hydrochloride and hydrocortisone in ear drops by high performance liquid chromatography

The aim of the study was to design and validate a reversed phase high performance liquid chromatography method for the separation and quantification of two active pharmaceutical ingredients (ciprofloxacin hydrochloride, hydrocortisone) and a preservative (benzyl alcohol) in ear drops. Effective separation of the examined compounds was achieved on a GraceSmart™ RP 18 column (150 mm × 4.6 mm, 5 μm) with gradient elution and a diode array detector. The total assay run time was 25 min. Analytical method validation assays were performed. Validation parameters used for the evaluation were: specificity, linearity, trueness, precision (repeatability and reproducibility), limit of detection and limit of quantitation. Results of the validation procedure (high recoveries, good standard deviations, no interfering peaks at the retention times corresponding to the analytes) confirm that the developed chromatographic method can be applied for routine analysis of ear drops.

Prediction of antimicrobial activity of imidazole derivatives by artificial neural networks

The main goal of our study is the analysis of data obtained from molecular modeling for a series of imidazole derivatives that possess strong antifungal activity. The research was designed to use artificial neural network (ANN) analysis to determine quantitative relationships between the structural parameters and anti-Streptococcus pyogenes activity of a series of imidazole derivatives. ANN in association with quantitative structure-activity relationships (QSAR) represents a promising tool in the search for drug candidates among the practically unlimited number of possible derivatives. In this work, a series of 286 imidazole derivatives presented as cationic three-dimensional structures was used. The activity was expressed as a logarithm of the reciprocal of the minimal inhibitory concentrations, log 1/MIC. Multilayer perceptron ANN was used for predictions of antimicrobial potency of new imidazole derivatives on the basis of their structural descriptors. The obtained correlation coefficient equaled 0.9461 for the learning set, 0.9060 for the validation set and 0.8824 for the testing set of imidazole derivatives. Hence, satisfactory and practically useful predictions of anti-Streptococcus pyogenes activity for a series of imidazole derivatives was obtained, supporting the future successful interpretation of QSAR analysis for those compounds.

Application of Rough Set Theory to Prediction of Antimicrobial Activity of Bis-Quaternary Imidazolium Chlorides

The paper investigates relationships between chemical structure, surface active properties and antibacterial activity of 70 bis-quaternary imidazolium chlorides. Chemical structure and properties of imidazolium chlorides were described by 7 condition attributes and antimicrobial properties were mapped by a decision attribute. Dominance-based Rough Set Approach (DRSA) was applied to discover a priori unknown rules exhibiting monotonicity relationships in the data, which hold in some parts of the evaluation space. Strong decision rules discovered in this way may enable creating prognostic models of new compounds with favorable antimicrobial properties. Moreover, relevance of the attributes estimated from the discovered rules allows to distinguish which of the structure and surface active properties describe compounds that have the most preferable and the least preferable antimicrobial properties.

Prediction of Antifungal Activity of Gemini Imidazolium Compounds

The progress of antimicrobial therapy contributes to the development of strains of fungi resistant to antimicrobial drugs. Since cationic surfactants have been described as good antifungals, we present a SAR study of a novel homologous series of 140 bis-quaternary imidazolium chlorides and analyze them with respect to their biological activity against Candida albicans as one of the major opportunistic pathogens causing a wide spectrum of diseases in human beings. We characterize a set of features of these compounds, concerning their structure, molecular descriptors, and surface active properties. SAR study was conducted with the help of the Dominance-Based Rough Set Approach (DRSA), which involves identification of relevant features and relevant combinations of features being in strong relationship with a high antifungal activity of the compounds. The SAR study shows, moreover, that the antifungal activity is dependent on the type of substituents and their position at the chloride moiety, as well as on the surface active properties of the compounds. We also show that molecular descriptors MlogP, HOMO-LUMO gap, total structure connectivity index, and Wiener index may be useful in prediction of antifungal activity of new chemical compounds.