Bioinformatics needs to adopt statistical thinking.

Editorial BIOINFORMATICS NEEDS TO ADOPT STATISTICAL THINKING Until a couple of years ago a typical question from my colleagues on the biological workbench would be phrased something like: " I have this sequence here and cannot find out anything about it. Can you help? ". From today's standpoint two things are remarkable about this question. First, it deals with only one sequence. Today, the question might be " I have got 2500 sequences... ". Secondly, even when using updated terms, the question is rarely asked. More typically, I now get asked questions like: " I have 17 hybridisations of such-and-such material versus an array of 10 000 genes. Can you help me interpret the data? ". This new type of question reflects the change of paradigm in molecular biology. When I was at the European Molecular Biology Laboratory as a doctoral student with a background in mathematics, among the first things I had to learn was to think in terms of experiments. I was extremely impressed by the care taken in design and setup of experiments and their controls. However, the typical experiment would result in a very concise output, for example a particular gel showing the binding of two components, or a microscopy image showing co-localisation of a molecule with some marker. Today's experiments are fundamentally different. The genome of an organism is sequenced with the goal of generating information to answer more than one question. But the sequence in itself is not the answer—not even the human sequence. Likewise, a micro-array experiment is not typically done with the goal of proving that a particular gene is up-regulated under certain conditions. Instead, the experiment generates a wealth of information that awaits interpretation. The transition from 'small science' to 'big science' has to bring about fundamental changes in our way of thinking about biological data. On a very general level, one can reason about hypothesis-driven research in contrast to hypothesis-free data generation in genomics. But it is only the entry point for the hypothesis that has changed. We now pose questions to this pool of data and these questions constitute our hypothesis. The fact that bioinformatics tools allow us to ask many different questions in a relatively short time does not make this type of research hypothesis free. We should, however, appreciate how different this procedure is in light of the tradition of molecular biology. Many experimentalists experience considerable frustration …