Artificial neural networks limitations

In many cases, structure elucidation with artificial neural networks is limited to backpropagation networks [113] and, is therefore performed in a supervised man-... 

The rather time- and cost-expensive preparation of primary brain microvessel endothelial cells, as well as the limited number of experiments which can be performed with intact brain capillaries, has led to an attempt to predict the blood-brain barrier permeability of new chemical entities in silico. Artificial neural networks have been developed to predict the ratios of the steady-state concentrations of drugs in the brain to those of the blood from their structural parameters [117, 118]. A summary of the current efforts is given in Chap. 25. Quantitative structure-property relationship models based on in vivo blood-brain permeation data and systematic variable selection methods led to success rates of prediction of over 80% for barrier permeant and nonper-meant compounds, thus offering a tool for virtual screening of substances of interest [119]. 

Traditional regression-type models have been linear and quadratic regression models. Linear and quadratic regression models unfortunately impose further constraints upon the nature of the process nonlinearity as such, these models are limited in the range of their applicability. A relatively new nonlinear regression-type model—the Artificial Neural Network (ANN)—is not as limited, and is worthy of additional discussion. 

Alany, R. G, Agatonovic-Kustrin, S., Rades, T., and Tucker, I. G. (1999), Use of artificial neural networks to predict quaternery phase systems from limited experimental data,/. Pharm. Biomed. Anal., 19(3-4), 443 152. 

In the previous chapter a simple two-layer artificial neural network was illustrated. Such two-layer, feed-forward networks have an interesting history and are commonly called perceptrons. Similar networks with more than two layers are called multilayer perceptrons, often abbreviated as MLPs. In this chapter the development of perceptrons is sketched with a discussion of particular applications and limitations. Multilayer perceptron concepts are developed applications, limitations and extensions to other kinds of networks are discussed. 

In most cases, the MFTA models are built using the Partial Least Squares Regression (PLSR) technique that is suitable for the stable modeling based on the excessive and/or correlated descriptors (under-defined data sets). However, the MFTA approach is not limited to the PLSR models and can successfully employ other statistical learning techniques such as the Artificial Neural Networks (ANN) supporting the detection of the nonlinear structure-activity relationships. ... 

An example of a non-covalent MIP sensor array is shown in Fig. 21.14. Xylene imprinted poly(styrenes) (PSt) and poly(methacrylates) (PMA) with 70 and 85% cross-linker have been used for the detection of o- and p-xylene. The detection has been performed in the presence of 20-60% relative humidity to simulate environmental conditions. In contrast to the calixarene/urethane layers mentioned before, p-xylene imprinted PSts still show a better sensitivity to o-xylene. The inversion of the xylene sensitivities can be gathered with PMAs and higher cross-linker ratios. As a consequence of the humidity, multivariate calibration of the array with partial least squares (PLS) and artificial neural networks (ANN) is performed, The evaluated xylene detection limits are in the lower ppm range (Table 21.2), whereas neural networks with back-propagation training and sigmoid transfer functions provide the most accurate data for o- and p-xylene concentrations as compared to PLS analyses. 

Zolotariou, R. Anwar, J. Modelling properties of powders using artificial neural networks and regression the case of limited data. J. Pharm. Pharmacol. 1998, 50 (Suppl)190. 

The major limitation of the simple perceptron model is that it fails drastically on linearly inseparable pattern recognition problems. For a solution to these cases we must investigate the properties and abilities of multilayer perceptrons and artificial neural networks. 

Fig. 4.19 Artificial neural network (ANN) and SUBSTRUCT classification of CNS activity. Result of the ANN classification of CNS activity (left) the correct classification rate is given as a function of the limit separating CNS-i- and CNS- molecules, respectively. For example, if the limit is 0.5 all molecules with a score <0.5 are classified as...

One commercially available sensor array analysis system, offered by Mosaic Industries [51], is Rhino , a microprocessor-based instrument with an array composed of discrete, resistive gas sensors. An artificial neural network processes sensor inputs and relates them to patterns established by training the instrument with gas components and mixtures of interest for a specific application. In principle, each system is customized for an application by the choice of sensors and the gas detection needs. Potential applications for this system are limited by the availability of suitable sensors and the complexity needed for discrimination. 

Between 1959 and 1960, Bernard Wildrow and Marcian Hoff of Stanford University, in the United States developed the ADALINE (ADAptive LINear Elements) and MADELINE (Multiple ADAptive LINear Elements) models. These were the first neural networks that could be applied to real problems. The ADALAINE model is used as a filter to remove echoes from telephone lines. The capabilities of these models were again proven limited by Minsky and Papert (1969). The period between 1969 and 1981 resulted in much attention toward neural networks. The capabilities of artificial neural networks were completely blown out of proportion by writers and producers of books and movies. People believed that such neural networks could do anything, resulting in disappointment when people realized that this was not so. 

---