Mirador

Mirador is a visualization tool I have created for exploratory analysis of complex tabular datasets that include mixture of discrete and continuous variables. Mirador started from a collaboration between the Sabeti Lab and Fathom Information Design. I have been using this tool in my own research to perform quick checks of correlation patterns between pairs of variables and univariate distributions.

My goal with Mirador is to create an “universal visualizer”, and by that I mean a tool capable to handle a wide range of datasets and allowing users to explore “all the data” at once with no prior knowledge of which patterns could be the most “interesting”. This goal poses a number of challenges though: how to consistenly represent (qualitatively and quantitatively) associations between variables of different types, and how to account for “false discoveries”, which are the result of looking at the data long enough until finding a significant correlation that’s not real but the result of chance (see P-Hacking).

Models, and more generally any kind of insights, derived from visual exploratory analysis will be affected by biases and significance over-estimation unless the false discovery rate in interactive visualization is accounted for. From Daniel Russo and James Zou in 2015: “Any data-exploration renders standard statistical theory invalid.” This is an ongoing area of research and I’m using Mirador as an testing platform to evaluate new methodologies to control for false discovery rate in interactive exploration.

We have also used Mirador in data analysis workshops as part of the African Center of Excellence for Genomics of Infectios Disease’s (ACEGID) training program at Harvard and the Broad institute:

ACEGID data analysis workshop with Mirador.

Mirador is open-source and installers are freely available for Windows, Mac, and Linux, as well as its source from GitHub. You can find tutorials and other resources in its homepage.