State-of-the-Art Neural Networks Applications in Biology

Introduction

An ANN is a system consisting of a large number of highly interconnected processing units, called neurons or nodes, which operate in unity to solve specific problems (Schalkoff, 1997). The structure and function of an artificial neuron resembles those of a biological neuron. In particular, a biological neuron is composed of dendrites, axons and a main body or soma. The dendrites accept signals from other neurons and transmit them to the soma. The axon receives signals from the cell body and transfers them through the synapse (a microscopic gap) to the dendrites of nearby neurons. In a similar way, a simple artificial neuron receives input signals, combines them to form an accumulated input, transfers them through a linear threshold gate and transmits the output signal to another neuron or to the environment. The connections between the nodes of an ANN are equivalent to the axons and dendrites of a biological neural network, the connection weights represent the conductivity of the synapses, while the threshold resembles the activity in the main body of the biological neuron (Jain et al., 1996).

Information processing is quite different between a conventional computing system and an ANN. The function of a conventional computing system is based on one or a few central processing units that can have access to an array of memory locations. In order for the processing unit to make a computation, instructions as well as the data they require are read from their memory addresses, the instructions are then executed and the results are stored in memory locations. The computational steps are sequential and deterministic, and the value of a given variable can be traced during all operations. In contrast, ANNs are massively parallel. An ANN is composed of many simple processing units instead of a few complex central processors, and responds to a pattern of input data presented to it. In addition, data is not stored in specific memory addresses, since information takes the form of the state of the network.

ANNs have proved to be particularly efficient computational systems due to a number of information processing characteristics they have. Specifically, their ability to learn allows them to adapt their internal structure in response to changing environment. Furthermore, their nonlinearity allows better fit of the ANN model to the data and their insensitivity to noise implies accurate prediction in cases when input data contain noise and measurement errors. In addition, the high parallelism that characterizes ANNs gives the ability for fast processing and tolerance to hardware failure (Basheer & Hajmeer, 2000). ANNs have have been applied successfully to a wide range of problems, such as pattern recognition, clustering, function approximation, forecasting, optimization, association and control problems.

In this paper, recent applications of ANNs in bioinformatics will be explored and ANN-based advances in the disease diagnosis of several diseases, in proteomics and genetics, will be presented. In addition, issues such as the role of ANNs in the establishment of an anti-malarial chemoinformatics platform, as well as the extension of neural networks using game theory, will be explored.