Neural networks model development

Log P can be used as an additional parameter, in combination with other descriptors. For example, neural network models developed by Liu and So and Goller et al use log P in combination with topological and quantum-chemical descriptors. Many methods do not use log R as a descriptor. These methods have been described in several reviews. However, there is a clear relationship between these two physicochemical properties, namely log P and aqueous solubility. 

Intended Use The intended use of the model sets the sophistication required. Relational models are adequate for control within narrow bands of setpoints. Physical models are reqiiired for fault detection and design. Even when relational models are used, they are frequently developed bv repeated simulations using physical models. Further, artificial neural-network models used in analysis of plant performance including gross error detection are in their infancy. Readers are referred to the work of Himmelblau for these developments. [For example, see Terry and Himmelblau (1993) cited in the reference list.] Process simulators are in wide use and readily available to engineers. Consequently, the emphasis of this section is to develop a pre-liminaiy physical model representing the unit. 

In this approach, connectivity indices were used as the principle descriptor of the topology of the repeat unit of a polymer. The connectivity indices of various polymers were first correlated directly with the experimental data for six different physical properties. The six properties were Van der Waals volume (Vw), molar volume (V), heat capacity (Cp), solubility parameter (5), glass transition temperature Tfj, and cohesive energies ( coh) for the 45 different polymers. Available data were used to establish the dependence of these properties on the topological indices. All the experimental data for these properties were trained simultaneously in the proposed neural network model in order to develop an overall cause-effect relationship for all six properties. 

Recently, Jung et al. [42] developed two artificial neural network models to discriminate intestinal barrier-permeable heptapeptides identified by the peroral phage display experiments from randomly generated heptapeptides. There are two kinds of descriptors one is binary code of amino acid types (each position used 20 bits) and the other, which is called VHSE, is a property descriptor that characterizes the hydrophobic, steric, and electronic properties of 20 coded amino acids. Both types of descriptors produced statistically significant models and the predictive accuracy was about 70%. 

Bodor, et al. [42] compare the use of artificial neural networks with regression analysis techniques for the development of predictive solubility models. They report that the performance of the neural network model is superior to the regression-based model. Their study is based on a training set of 331 compounds. The model requires a diverse set of molecular descriptors to account for the structural variety in the training compounds. 

Using artificial neural networks to develop calibration models is also possible. The reader is referred to the literature [68-70] for further information. Neural networks are commonly utilized when the data set maintains a large degree of nonlinearity. Additional multivariate approaches for nonlinear data are described in the literature [71, 72],... 

For the development of neural network models simulated process operation data from 50 batches with different feeding profiles were generated using the mechanistic model of the process. In each batch, the batch duration is divided into 10 equal stages. Within each stage, the feed rate is kept constant. The control policy for a batch consists of the feed rates at these 10 stages. 

The multiple linear regression models are validated using standard statistical techniques. These techniques include inspection of residual plots, standard deviation, and multiple correlation coefficient. Both regression and computational neural network models are validated using external prediction. The prediction set is not used for descriptor selection, descriptor reduction, or model development, and it therefore represents a true unknown data set. In order to ascertain the predictive power of a model the rms error is computed for the prediction set. 

The regression model, the neural network model, and a group contribution model developed by Gao [1] were then compared. Gao used 16 structural groups to develop a model. Table 3 provides a comparison predictive power of the three approaches. The 16-descriptor group contribution model gave the smallest overall rms error of 0.40. The neural network model using only 5 descriptors gave an rms error of 0.41. 

Ktihne et al. [105] used MP and 23 structural parameters to correlate the vapor pressure of 1838 hydrocarbons and halogenated hydrocarbons. The neural network model was built using 1200 compounds and provided a mean absolute error of 0.13 log units for the test set of 638 compounds. It should be noted that the authors predicted vapor pressure as a function of the temperature and total number of data points for model development was 8148. The MP and BP are also used in MPBPVP program developed by Syracuse Research Inc [101]. 

Plasma area under the concentration—time curves (AUCs) of 57 NCEs were determined following oral cassette administration (5—9 NCEs/cassette) to mice. Physicochemical properties [such as, molecular weight, calculated molar refractivity, and calculated lipophilicity (clogP)] and molecular descriptors [such as presence or absence of N-methylation, cyclobutyl moiety, or heteroatoms (non-C,H,0,N)] were calculated or estimated for these compounds. This structural data, along with the corresponding pharmacokinetic parameters (primarily AUC), were used to develop artificial neural network models [8]. These models were used to predict the AUCs of compounds under synthesis [10]. This approach demonstrates that predictive models could be developed which potentially predict in vivo pharmacokinetics of NCEs under synthesis. Similar examples have been reported elsewhere [11—13]. 

Thermal processing of food by conduction heating Simultaneous minimization of surface cook values (i.e. maximization of final product quality) and minimization of processing time. GA An artificial neural network model was developed based on simulated data from the first principles model, and then used in optimization. Chen and Ramaswamy (2002)... 

Dielectric barrier discharge reactor for conversion of methane and CO2 into synthesis gas and C2+ hydrocarbons Three cases (a) maximization of metiiane conversion and C2+ selectivity, (b) maximization of methane conversion and C2+ yield, and (c) maximization of methane conversion and H2 selectivity. Weighted sum of squared objective functions method along with GA An artificial neural network model of the process was developed based on experimental data, and then used for optimization. Istadi and Amin (2006)... 

In this paper, a new computer-aided technique was presented, with which the experimental procedure of developing catalysts is scheduled sequentially. In each sequential step the neural networks model and multi-objective optimization are used to determine optimal design for the next experiment. The sequential method proved very efficient in developing catalysts for propane ammoxidation to acrylonitrile. And the yield of acrylonitrile corresponding to the best catalyst was up to 58.9%. 

---