All of my research has been based on computation, starting with my Master's thesis (1989) on the dynamics of colloidal suspensions. Back then, computational physics concentrated on simple systems, and scientists usually wrote their own simple software from scratch. When I started to work on biomolecular simulations five years later, I discovered the sad state of non-trivial scientific software, written by generations of PhD students and postdocs who learned Fortran "on the job" and had the prime goal of getting their research project done with a minimum of programming effort.
I started looking for better tools for scientific software, discovered the Python language, and joined the small group of researchers who were developing the "Numerical Python" package (the precursor of today's NumPy with the goal of doing computational science in Python. When an HFSP grant gave me the freedom to choose the tools I like, I decided to write a Python library for biomolecular simulation and to publish it as an Open Source package. It's called the Molecular Modeling Toolkit (MMTK), and it has become popular for developing molecular simulation methods. When distributed version control systems appeared, I adopted Mercurial for MMTK development, which is now hosted on Bitbucket.
MMTK has been the central tool for most of my own research work on protein structure and dynamics, in particular for the development of Elastic Network Models. I try to publish the MMTK-based Python scripts underlying my publications (here's an example), so that others can reproduce the results and apply the techniques to their own problems. Others have adopted MMTK for building their own tools, examples are the visualization program Chimera, the normal-mode Web application WEBnmS, and the MD analyis package nMOLDYN. All this became possible because MMTK is published as Open Source software.
Like everything related to computing technology, scientific computing has been evolving at a rapid pace and this process is likely to continue. The biggest challenge I see today is the integration of software into the process of doing science. We tend to consider software as the equivalent of some experimental apparatus: a tool that does a well-defined job, which the scientists needs to understand of course, but whose construction and detailed workings can be left to specialists. However, computing has been growing in importance for research to the point that it ranks equal to mathematics and to many experimental techniques. For more and more research studies, the "methods" part is nothing but algorithms. We need to make sure that scientists understand these algorithms, but we also need to make sure that the highly optimized versions of these algorithms that are implemented in software are indeed equivalent to the simple versions that are presented in the papers. A question that we need to ask ourselves constantly is: given that software is known to have bugs, why should anyone believe that our computational results are right?
The Reproducible Research movement is making an important contribution to this development by encouraging more openness and transparency in computational science, and I am adopting reproducible research for my own work in molecular simulation. Since existing tools did not support large binary datasets and projects requiring multiple computing platforms, I am developing my own toolset called ActivePapers. Molecular simulation has the additional problem of a lack of flexible and documented file formats for publishing and archiving detailed and machine-readable descriptions of simulations. I am working on that aspect as well in the MOSAIC project.
Still, the biggest obstacle for more reliable computational science is not technology, but the lack of problem awareness in the scientific community and the lack of computational compentence among practicing scientists. And that's why I joined Software Carpentry as an instructor. I am convinced that these bootcamps are useful, but they are also fun, and a good way to meet scientists from different fields who care about the same problems.comments powered by Disqus