ABSTRACT
Neural network methods are recognized as holding great promise for advanced understanding of biological systems, associated phenomena, and related processes. For example, the ability of neural networks to learn arbitrary complex functions from large amounts of data without the need for predetermined models makes them an ideal tool for protein structure prediction (Wood and Hirst, 2005). Others have utilized the powerful classi€cation capabilities of neural networks for the analysis and unsupervised classi€cation of electron microscopy images of biological macromolecules (Pascual-Montano et al., 2001). Moreover, they can aid in acquiring accurate knowledge of quantitative structure-activity relationships (e.g., Jalali-Heravi et al., 2008; Lara et al., 2008), a process by which chemical structure is quantitatively correlated with biological activity (or chemical reactivity). Regardless of the application area, neural networks can offer sophisticated modeling tools when faced with a profusion of data, even if there is an inadequate understanding of the biological system under investigation. Yet, despite the increased intricacy in model representations, in practice it has been demonstrated that anticipated improvements have frequently not lived up to expectations. Having said that, the applications presented in this chapter provide the reader with encouraging detail in regard to the use of neural networks in contemporary biological research efforts, and comment directly on ways in which developed models can be enhanced. Model framework, development, validation, and application concepts covered comprehensively in previous chapters will provide context for the coverage of these investigations.
