I do research in the field of computational biology. It's a broad field.
What I don't work on is the best known area of computational biology, which is called bioinformatics. Bioinformatics is the best known area of computational biology, and involves the use of computers to store, search, and draw statistical conclusions from large amounts of biological information, usually genomic data.
What I do work on is computational modeling of biological processes. Specifically, I use computers to model populations of organisms with a genotype and a developmental process (the mapping from the genotype to the phenotype). I am interested in evolvability, defined as, "the likelihood that the mapping from the genotype to the phenotype will produce adaptive phenotypic variation when the genotype is subjected to common types of genetic variation (such as mutation and recombination)."
Some people, including me, suspect that there is indirect selection for evolvability over the long term, i.e., that there is selective pressure for developmental processes that tend to produce more, rather than less, adaptive variation. If that were true, it would help explain how very complicated artfacts - like you and me - could be generated by evolution. Such artifacts are, without exception, suspiciously hierarchical and modular, which makes one wonder whether this kind of organization is a prerequisite to evolution having any reasonable chance of generating complex artifacts.
Unfortunately, it's very difficult to study evolvability in the biological context. The main problem is that big changes to the genotype-phenotype map happen over very long time scales, e.g., 1e5 to 1e8 years. I expect to live only on the order of 1e2 years - and I'm already almost 4e1. There is some fossil evidence and some genomic evidence you can consider, but there's nothing quite as informative as doing an experiment.
Another practical problem is that the genotype-phenotype map in living things is damned complicated! Much of it remains beyond our understanding at present.
Happily, one can construct computer models of populations of organisms with an explicit, evolvable developmental process. Some biologists seem to doubt you can really draw conclusions about living things this way, but biology is not magic; in theory, one can identify the salient features of any physical (as opposed to magical) system and study them with a computer model (or with mathematics if they are not too complicated). And in fact, computer modelling is starting to yield solid results regarding the evolution of evolvability and the evolution of robustness.
The original reason I became interested in evolvability is that I want to evolve computer programs and other software objects. When I tell people I meet at cocktail parties, "I want to evolve computer programs,", I get a lot of quizzical looks. Even from biologists, which surprised me! "How can a computer program evolve?" If you have a population of artifacts that replicate with heritable variation, and the copies are subjected to differential selection (some of the variation is adaptive), then adaptive evolution will tend to occur (in large enough populations that drift doesn't overwhelm it). The artifacts can be bacteria, or people, or computer programs.
One thing that makes people question the idea of evolving a computer program is that they imagine taking a page of working computer code and randomly applying "mutations" to it, by changing letters. This has in fact been tried, but it doesn't work very well. Intuitively, one would expect this to create a lot of non-adaptive variation, which it does. The genetic representation is very important to evolvabilty; and, interestingly, the genetic representation itself can useful evolve, if you let it.
There are two things stopping evolutionary computing from being widely applied today: 1) people haven't figured out how to manipulate evolvabilty, so they can't evolve really complex artifacts (i.e., that couldn't be created more easily by another method, such as hiring a human programmer); and 2) we haven't thrown enough computing horsepower at the problem. I think that people are going to figure out how to manipulate evolvability in silico - I hope to be a part of the effort! Folks in the chip industry say that speeding up single CPUs is getting hard, and that most of the performance gains in the future will come from increased parallelism. Although some programs are inherently serial, and can't be sped up by parallelization, evolutionary computing parallelizes very easily. This means it will become relatively much cheaper than it is today. I would bet that within ten years (I write this in March, 2007) evolutionary computing will become a mainstream technique for doing some things that we have no idea how to do now. To quote Johnny Rotten of the Sex Pistols: "Don't know what I want / But I know how to get it!"