Artificial neural networks: introduction and some examples
by
N. Pencheva
Bulg. Acad. Sci., Institute of Physiology, Sofia
Coauthors: P. Milanov

Many aspects of brain function could be modelled using a direct network of neurones, which co-ordinate their firing. Any neurone fires if a weighted sum of the inputs to it from other neurones exceeds its threshold. The synaptic weight between any two neurones indicates the contribution, which the firing of the first neurone makes to the total input of the second neurone. Artificial neural networks, composed of varying numbers of neurones (organized in layers), with different sets of weights and different connection architectures, could represent a wide range of logical functions and possess considerable potential for modelling some mental activities. There are two distinct phases of operation of an artificial neural network: the learning or training stage and the operation stage.

To illustrate some of these ideas we analyze and comment neural networks taken from literature, which have proven their usefulness as tools for drug design.

Various types of artificial neural networks have been used to turn a blind search for novel drug-like molecules into an informed search and to select well-defined subsets of compounds from the accessible chemical space. The following tasks in drug design can be performed by artificial neural networks: (i) classification of large data sets by self-organizing networks; (ii) further nonlinear modelling of quantitative structure-activity relationships my multi-layered feedforword networks, self-organizing networks or hybrid architectures; and (iii) prediction of molecular properties.

Following this general scheme we comment successful examples leading to activity-enriched compound collections. The present material helps for better understanding of: (i) the important contribution of the artificial neural networks in drug-design process; and (ii) the dialogue between the disciplines such as mathematics, biochemistry, cognitive sciences etc.

Date received: August 1, 2001