patrick fuller's blog

home about

Data Management for Scientists

17 Jul 2012

I've spent the last few months working with the folks at Codecademy to create a tutorial targeted at experimentalists. It covers what I have found to be the most commonly-encountered-but-easily-solvable problem in experimental science: data management. Especially with older machines, data is often generated in non-ideal formats. The data might present itself in obscure text files, or maybe you only need every nth datapoint. Regardless, the usual solution that I've seen is for a scientist to spend hours manually copying and pasting important information. Not only is this dangerous - translational errors are a real thing - but it's miserably boring.

This issue came up in the work behind one of my earlier papers. In one set of experiments, we wanted to view the motion of a single dye molecule in various metal-organic framework crystals. Amazingly, this task can be accomplished through an approach called fluorescence correlation spectroscopy, or FCS. This method can monitor the fluorescence intensity of a small volume (think sub-micron) over time, providing information on flow. In a very dilute system, the approach can monitor the volume around single molecules, which allows us to extrapolate data on different time scales - we can observe Brownian motion, adsorption kinetics, and diffusion.

The resultant data of a set of experiments comes out looking like this large text file (I don't know if I have rights to the raw data, so I've mangled the values. I have, however, attempted to keep the underlying trends the same.). We needed to autocorrelate the data and fit to models to ascertain diffusion and adsorption coefficients, but we obviously couldn't do that in its current form. We moved the data to Excel using the approach described in the Codecademy tutorial, and were then able to fit our models. Victory.

If anyone has any questions or comments about the Codecademy tutorial or the underlying FCS experiments, feel free to ask!