Beautiful Data

Closing the Loop: Visualizations to Suggest New Experiments

As noted earlier, data from experiments can be utilized in a variety of ways, ranging from visualization to modeling. These activities are useful and can provide insight into the physical problem at hand. However, our main aim is to use the modeling and analysis to inform the design of new experiments. As the crowdsourcing effort expands, it is important to consider possible experiments and prioritize these, particularly if the ultimate aim is to enable interested, but not necessarily experienced, researchers to take part. Such computational prioritization is very useful in many scenarios, where resources (financial, material, time) are limited and all possible experiments cannot be carried out. In the case of solubility, an experimentalist might ask, "Given the compounds tested so far, which ones should we do next?" Visualization of the data can be both compelling and provide a good guide to the best choice of the next experiment given the resources available. This enables a cyclical relationship between experiment and computation, making optimal use of both the experimentalist's and analyst's skills.

To identify which compounds we, or anyone else, should test next, we need a way of understanding where in a "chemical space" each of the compounds we have already tested lie. Then it will be possible to identify empty parts of that space in our data set, correlate that with specific molecules that lie in those spaces, and ...