Principles of Plant Genetics and Breeding

242 CHAPTER 14
position or combination of positions discriminate one subset from the others? Using a number of statistical techniques (e.g.,
multivariate analysis, group theory), they were able to develop a model for how tRNA sequences allowed the aaRS enzymes to
identify the correct tRNA molecules and reject the incorrect ones (McClain 1995). Experimental verification (started prior to the
computational work in one case) of these results were published subsequently (Hou & Schimmel 1988; McClain & Foss 1988).
Additional use of the multiple sequence alignment
In addition to the above analysis of a well-crafted multiple sequence alignment, there are two additional areas that can use
the information contained within a multiple sequence alignment. The first is the creation of a phylogenetic tree for evolutionary
studies. The second is to allow more sensitive database searches using a representation that incorporates the pattern of substitution
seen in the multiple sequence alignment to allow researchers to find more highly diverged homologous sequences.
Pattern identification
Pattern identification has been developed to identify small, unique sections of the several unaligned sequences. Often, these
contiguous regions of conserved residues, called motifs, are important for molecular interactions, such as regulatory regions or
binding sites. Thus, motifs are often essential for the correct functioning of the molecule.
The classic example of pattern identification is to collect DNA sequences from the region just upstream (on the 5′ side) of the
coding region of a gene and examine these for a conserved pattern of nucleotides (Sadler et al. 1983) involved in regulating the
transcription of the genes. What distinguishes this problem from global multiple sequence alignment is that outside the conserved
patterns there is no expectation that the sequence is conserved and thus alignable.
Modern pattern identification programs (Bailey & Elkan 1994) make use of a modern statistical process designed to deal with
the fact that we do not know where the patterns are located (expectation maximization) and a sophisticated sampling routine
(stochastic sampling) that reduces the number of combinations that must be tried.
Other techniques
As the biomedical sciences have expanded their repertoire of research methods and the kinds of data that can be collected, the
field of bioinformatics has created techniques for dealing with these new kinds of data. The advent of the complete genome
sequences for many organisms has been accompanied by software to allow the manipulation, annotation, analysis, and comparison
of these large sequences. Complex mathematical models of genes try to find and identify all of the genes in each genome
(Rogic et al. 2001).
Techniques have been developed to measure the change in expression for cDNA (microarrays) or the amount of proteins in
cells over time or between mutants and wild types. It is not unusual for a research group to monitor thousands of molecules simultaneously;
looking for either increases or decreases in the relative levels between the standard and the state under investigation.
These large-scale experiments are being analyzed with a number of statistical techniques (Wetzel et al. 2000) such as analysis of
variance, which produces a statistical model of the changes observed (Kerr & Churchill 2001). Other researchers are using multivariate
statistical techniques to identify which molecules vary their presence in a coordinated manner in response to changing
conditions. Interestingly, a number of these techniques were first developed many years ago to study the factors influencing crop
growth.
Conclusion
Ultimately, though, the field of bioinformatics does have some general themes that should continue to run throughout it in the
future. First, the bioinformaticist tool chest is not complete – the tool chest of tomorrow will have only minimal relationship to
today’s set of tools, with better and more sensitive tools continuing to be developed. Second, the numbers and types of databases
of experimental data will continue to expand at an alarming rate. The majority of the databases will be developed to describe one
type of experimental data, like sequence data or microarray data, with only minimal references or consistencies (vocabulary) to
the other databases. Third is that diverse data must be integrated across ranges of scale, both temporal and spatial. For example, a
single point mutation in a mouse might cause kidney deformations that result in blood chemistry being incorrect. Thus, you have
a single point mutation causing effects at the cellular and organ levels. Biological scientists must learn the techniques necessary to
manage and make use of the new data resources that their research is creating.
References
Bailey, T.L., and C. Elkan. 1994. Fitting a mixture model by expectation maximization to discover motifs in biopolymers.
In: ISMB-94: Proceedings of the Second International Conference on Intelligent Systems for Molecular Biology (Altman, R.,
D. Brutlag, P. Karp, R. Lathrop, and D. Searls, eds), pp. 28–36. AAAI Press, Menlo Park, CA.
Crick, F.H.C. 1957. The structure of nucleic acids and their role in protein sysnthesis. In: Biochemical Society Symposium, No. 14
(Crook, E.M., ed.), pp. 26–36. Cambridge University Press, Cambridge, UK.