Which craft is best in bioinformatics?

'Silicon-based' biology has gathered momentum as the world-wide sequencing projects have made possible the investigation and comparative analysis of complete genomes. Central to the quest to elucidate and characterise the genes and gene products encoded within genomes are pivotal concepts concerning the processes of evolution, the mechanisms of protein folding, and, crucially, the manifestation of protein function. Our use of computers to model such concepts is limited by, and must be placed in the context of, the current limits of our understanding of these biological processes. It is important to recognise that we do not have a common understanding of what constitutes a gene; we cannot invariably say that a particular sequence or fold has arisen via divergence or convergence; we do not fully understand the rules of protein folding, so we cannot predict protein structure; and we cannot invariably diagnose protein function, given knowledge only of its sequence or structure in isolation. Accepting what we cannot do with computers plays an essential role in forming an appreciation of what we can do. Without this understanding, it is easy to be misled, as spurious arguments are often used to promote over-enthusiastic notions of what particular programs can achieve. There are valuable lessons to be learned here from the field of artificial intelligence, principal among which is the realisation that capturing and representing complex knowledge is time consuming, expensive and hard. If bioinformatics is to tackle biological complexity meaningfully, the road ahead must therefore be paved with caution, rigour and pragmatism.