Computational Systems Biology: solving problems rather than looking for problems

Biology and its concrete cases must drive the development of methods and software tools to enhance research capacity and quality. Unfortunately too many times I've seen computational approaches designed and tested on fake data sets or toy artificial problems for then trying to find out there the right application for the elegant method just developed. The outcome of this practice is a huge ecosystem of methods and tools in the scientific domain that is by far much larger than what is really needed to face the challenges posed by system biology.

At COSBI, we learned by doing how much more effective is developing methods and tools to solve concrete problems -- i.e., being driven by biology in designing computational approaches. In the early days at COSBI, 12 years ago, we experimented also the wrong way working -- i.e., developing methods that we thought could help biologists in their everyday work -- and the results were not so exciting. We developed amazing technologies for modelling and simulating biological systems, but then we realized that it was too difficult to effectively communicate with biologists. The technology was very elegant and flexible, but not efficient enough to manage systems of the size of practical applications. Of course, we re-used all that experience in the definition and implementation of the simplest graphical language ever to model systems together with biologists. We also re-used our initial development to equip it with the fastest stochastic simulation algorithm to approach industrial-size systems. But this could have been done in much less time, if we would have been driven from the very beginning by concrete biological problems.

Summing up, developing nice methods and then saying, look, I have the best computational method ever, give me a problem where my method apply is an investment of time and resources that is going to fail. The right strategy I suggest is: give me a problem, I check if a solution can be obtained by orchestrating in the right way what is already available, and, only as a last chance, eventually I'm going to develop a new computational method if really needed.

?