Visualizing uncertainty in biological expression data

Abstract

Expression analysis of ∼omics data using microarrays has become a standard procedure in the life sciences. However, microarrays are subject to technical limitations and errors, which render the data gathered likely to be uncertain. While a number of approaches exist to target this uncertainty statistically, it is hardly ever even shown when the data is visualized using for example clustered heatmaps. Yet, this is highly useful when trying not to omit data that is "good enough" for an analysis, which otherwise would be discarded as too unreliable by established conservative thresholds. Our approach addresses this shortcoming by first identifying the margin above the error threshold of uncertain, yet possibly still useful data. It then displays this uncertain data in the context of the valid data by enhancing a clustered heatmap. We employ different visual representations for the different kinds of uncertainty involved. Finally, it lets the user interactively adjust the thresholds, giving visual feedback in the heatmap representation, so that an informed choice on which thresholds to use can be made instead of applying the usual rule-of-thumb cut-offs. We exemplify the usefulness of our concept by giving details for a concrete use case from our partners at the Medical University of Graz, thereby demonstrating our implementation of the general approach.

LineUp: Visual Analysis of Multi-Attribute Rankings

“…It is even possible to go below the single pixel limit by using overplotting. This aggregation of multiple values to a single pixel introduces visual uncertainty [11]. While details such as outliers will be omitted, major trends will remain visible.…”

Conceptualizing Visual Uncertainty in Parallel Coordinates

“…On the human side, we have to take perception into account as well; uncertainty due to perception has been discussed by Russell and Norvig [RNC*95]. Holzhüter et al [HLS*12] describe uncertainty in visualization and differentiate between input and output uncertainty. Relating the information visualization pipeline to the communication channel as discussed below, we choose encoding and decoding uncertainty to be the topmost classifying schemes.…”

Taggle: Combining overview and details in tabular data visualizations

“…iHAT 50 aggregates amino acid sequences and associated metadata using the most frequent category or the average to represent aggregated items, depending on the data type. Holzhü ter et al 51 use the average for numerical values for aggregates. Both techniques employ transparency to communicate fidelity (the higher the variation in a cell, the higher the transparency), but neither addresses fidelity well.…”

“…To achieve this, Taggle decreases the height of items until the whole table fits on the screen, or until each item has a height of a single pixel, as lower values would introduce uncertainty due to interpolation artifacts. 51 Aggregated groups are shown using a fixed height. Overview mode is a complementary strategy to aggregation: it is useful to get an idea about the distribution of the data in the columns and does not require that meaningful groups are defined.…”