Modelling and prediction of retention in high-performance liquid chromatography by using neural networks

Summary Multi-layer feed-forward neural networks trained with an error back-propagation algorithm have been used to model retention behaviour of liquid chromatography as a function of the composition of the mobile phases. Conventional hydro-organic and micellar mobile phases were considered. Accurate retention modelling and prediction have been achieved using mobile phases defined by two, three and four parameters. With micellar mobile phases, the parameters involved included the concentrations of surfactant and organic modifier, pH and temperature. It is shown that neural networks provide a competitive tool to model varied inherent nonlinear relationships of retention behaviour with respect to the mobile phase parameters. The soft models defined by the weights of the networks are capable of accommodating all types of linear and nonlinear relationships, neural networks being specially useful when the relationships between retention behaviour and the mobile phase parameters are unknown. However, to train neural networks more experimental points than with hard-modelling methods are required, hence the use of the networks is recommended only for those cases where adequate theoretical or empirical models do not exist.     

Artificial neural networks study of the catalytic reduction of resazurin: stopped-flow injection kinetic-spectrophotometric determination of Cu(II) and Ni(II)

An artificial neural network (ANN) procedure was used in the development of a catalytic spectrophotometric method for the determination of Cu(II) and Ni(II) employing a stopped-flow injection system. The method is based on the catalytic action of these ions on the reduction of resazurin by sulfide. ANNs trained by back-propagation of errors allowed us to model the systems in a concentration range of 0.5-6 and 1-15 mg l⁻¹ for Cu(II) and Ni(II), respectively, with a low relative error of prediction (REP) for each cation: REP_Cu(II) = 0.85% and REP_Ni(II) = 0.79%. The standard deviations of the repeatability (s_r) and of the within-laboratory reproducibility (s_w) were measured using standard solutions of Cu(II) and Ni(II) equal to 2.75 and 3.5 mg l⁻¹, respectively: s_r[Cu(II)] = 0.039 mg l⁻¹, s_r[Ni(II)] = 0.044 mg l⁻¹, s_w[Ni(II)] = 0.045 mg l⁻¹ and s_w[Ni(II)] = 0.050 mg l⁻¹. The ANNs-kinetic method has been applied to the determination of Cu(II) and Ni(II) in electroplating solutions and provided satisfactory results as compared with flame atomic absorption spectrophotometry method. The effect of resazurin, NaOH and Na₂S concentrations and the reaction temperature on the analytical sensitivity is discussed.     

Development of a chemical vapor sensor using piezoelectric quartz crystals with coated unusual lipids

Piezoelectric quartz crystal sensors were developed using lipids with various properties for highly sensitive detection of chemical vapors. Lipids with varying lengths of alkyl chains were coated onto 10 MHz AT-cut quartz crystal resonators and the response of these modified crystals to chemical vapors were measured. It was shown that hydrophilic compounds, such as ethanol and methanol, could be recognized efficiently by lipids having shorter alkyl chains, whereas lipids with longer alkyl chains showed affinity to more hydrophobic vapors, such as toluene, hexane and cyclohexane. Frequency changes caused by adsorption of alcohols could be enhanced when cholesterol was co-immobilized in the lipid layer. To confirm the assumption that the sensor-response might be affected by the properties of lipids derived from acyl chains, we have examined the effects of two types of newly synthesized unusual lipids on sensor response. When lipids having one triple bond each at different positions on their alkyl chains were coated onto quartz crystals separately, lower responses were observed compared to responses obtained for a sensor with immobilized, saturated phosphatidylcholine. Lipids containing -branched acyl chains, however, showed good affinity for organic vapors, and sensor responses improved 4–5-fold. Moreover, these sensors were shown to have sensitivity of the same order as the humans' sense of smell (10⁻⁵–10⁻⁶ w/w in liquid paraffin) when measured using standard odorants (isovaleric acid, skatole, etc.) for an olfactometry established in Japan.     

Artificial red cells with crosslinked hemoglobin membranes

Artificial cells containing concentrated hemoglobin (Hb) solution were prepared by interfacial polymerization of Hb with glutaraldehyde (GA) in liquid membrane capsules (LMC). A solution containing 30% of Hb was emulsified in mineral oil as red cell-size microdroplets, and this emulsion was dispersed in an aqueous phase containing glutaraldehyde to form LMC. The LMC were semipermeable templates that held the microdroplets of Hb in suspension while GA diffused through the oil to the microdroplet surfaces. The GA crosslinked Hb at the surface of each microdroplet to form an artificial red-cell membrane encapsulating Hb solution. A water-soluble surfactant was used to eject the cells from the LMC and suspend them in saline. Several surfactants were evaluated. Cell size was controlled by agitation speed during preparation of the original emulsion. Cells of 2.52 = ±1.69 μm were prepared. The encapsulated Hb retained capacity to bind and release O₂. The cells had aP ₅₀ of 8.9 torr (1200 Pa) and a capacity of 0.55 cc O₂/g of total Hb, indicating that the crosslinked portion of the Hb did not contribute to O₂ transport.     

自组织神经树用于判别氟化物非晶态形成条件

1人工神经网络──自组织神经树模型人工神经网络（ANN）是八十年代中期迅速兴起的一门非线性科学．它力图模拟人脑的一些基本特性，如自适应性、自组织性、容错性等，已在模式识别、数据处理及自动化控制等领域得到了初步应用，取得了相当好的效果［1，2］．1993年，Tao－Li等提出了自组织神经树网络．它是一个多层树状网络（见图1），每个输入节点与所有神经树的节点（神经元）通过权W相联系，实现对输入信号的非线性降维映射．映射中保持拓扑不变性，即把拓扑意义下相似的输入（即在高维空间中距离较近的输入向量）映射到相同子树的节…     

Toward a new generation of therapeutics

Scientific evidence in the prevention and treatment of various disorders is accumulating regarding probiotics. The health benefits supported by adequate clinical data include increased resistance to infectious disease, decreased duration of diarrhea, management of inflammatory bowel disease, reduction of serum cholesterol, prevention of allergy, modulation of cytokine gene expression, and suppression of carcinogen production. Recent ventures in metabolic engineering and heterologous protein expression have enhanced the enzymatic and immunomodulatory effects of probiotics and, with time, may allow more active intervention among critical care patients. In addition, a number of approaches are currently being explored, including the physical and chemical protection of cells, to increase probiotic viability and its health benefits. Traditional immobilization of probiotics in gel matrices, most notably calcium alginate and κ-carrageenan, has frequently been employed, with noted improvements in viability during freezing and storage. Conflicting reports exist, however, on the protection offered by immobilization from harsh physiologic environments. An alternative approach, microencapsulation in “artificial cells,” builds on immobilization technologies by combining enhanced mechanical stability of the capsule membrane with improved mass transport, increased cell loading, and greater control of parameters. This review summarizes the current clinical status of probiotics, examines the promises and challenges of current immobilization technologies, and presents the concept of artificial cells for effective delivery of therapeutic bacterial cells.     

Sequential injection system with higher dimensional electrochemical sensor signals Part 2. Potentiometric e-tongue for the determination of alkaline ions

An intelligent, automatic system based on an array of non-specific-response chemical sensors was developed. As a great amount of information is required for its correct modelling, we propose a system generating it itself. The sequential injection analysis (SIA) technique was chosen as it enables the processes of training, calibration, validation and operation to be automated simply. Detection was carried out using an array of potentiometric sensors based on PVC membranes of different selectivity. The diluted standard solutions needed for system learning and response modelling are automatically prepared from more concentrated standards. The electrodes used were characterised with respect to one and two analytes, by means of high-dimensionality calibrations, and the response surface of each was represented; this characterisation enabled an interference study of great practical utility. The combined response was modelled by means of artificial neural networks (ANNs), and thus it was possible to obtain an automated electronic tongue based on SIA. In order to identify the ANN which provided the best model of the electrode responses, some of the network's parameters were optimised and its usefulness in determining NH₄⁺, K⁺ and Na⁺ ions in synthetic samples was then tested. Finally, it was used to determine these ions in commercial fertilisers, the obtained results being compared with reference methods.     

A multineuron interaction model for neural networks

A multineuron interaction model (RS model) with an energy function given by the product of the squared distances in phase space between the state of the net and the stored patterns is studied in detail within a mean-field approach. Two limits are considered: when the patterns and antipatterns are stored (as in the Hopfield model), PAS case, and when only the patterns are taken into account, OPS case. TheT=0 solutions for the proper memories are exactly obtained for all finite values of, as a consequence of the energy function: whenever one of the overlaps is exactly one the corresponding equations decouple and no configuration average is required. Special interest is focused on the OPS situation, which presents a peculiar phase space topology. On the other hand, the PAS configuration recovers the Hopfield model in the appropriate limit, while keeping associative memory abilities far beyond the critical values of other models when the full Hamiltonian is considered.