GenoREC: A Recommendation System for Interactive Genomics Data Visualization

Abstract

Interpretation of genomics data is critically reliant on the application of a wide range of visualization tools. A large number of visualization techniques for genomics data and different analysis tasks pose a significant challenge for analysts: which visualization technique is most likely to help them generate insights into their data? Since genomics analysts typically have limited training in data visualization, their choices are often based on trial and error or guided by technical details, such as data formats that a specific tool can load. This approach prevents them from making effective visualization choices for the many combinations of data types and analysis questions they encounter in their work. Visualization recommendation systems assist non-experts in creating data visualization by recommending appropriate visualizations based on the data and task characteristics. However, existing visualization recommendation systems are not designed to handle domain-specific problems. To address these challenges, we designed GenoREC, a novel visualization recommendation system for genomics. GenoREC enables genomics analysts to select effective visualizations based on a description of their data and analysis tasks. Here, we present the recommendation model that uses a knowledge-based method for choosing appropriate visualizations and a web application that enables analysts to input their requirements, explore recommended visualizations, and export them for their usage. Furthermore, we present the results of two user studies demonstrating that GenoREC recommends visualizations that are both accepted by domain experts and suited to address the given genomics analysis problem. All supplemental materials are available at https://osf.io/y73pt/.

Fig. 1.

Top: GenoREC maps data and task specifications (A) to appropriate visualizations. In this figure, the knowledge-based recommendation (B) shows the component-wise model of GenoREC and the subsequent decisions made at each step. Based on the recommendation model, GenoREC generates and recommends an appropriate visualization to the user (C). Through the recommendation, GenoREC allows the user to avoid a wide range of similar but sub-optimal visualization options (D) given the data and task. Bottom: An overview of GenoREC’s system components and their interactions to generate output visualizations.

Fig. 2.

Visual overview of a genome and genomic features: point, segment, sparse, and contiguous.

Fig. 3.

Progression of GenoREC’s recommendation. In A, user has specified a BED file and a VCF file with categorical data and GenoREC recommends linear and circular layouts. In B, user adds a BIGWIG file with quantitative data and GenoREC switches to a linear recommendation.

Fig. 4.. User Interface of GenoREC.

The interface consists of two main panels. Left: a domain-centric data and task elicitation interface. The data and task specification panel contains two parts data description and tasks. Right: a visualization gallery that displays recommends visualizations.

Fig. 5.

Decision Matrix for the Encoding component (C1). Rows represent the visual encodings, and the columns represent the factors that affect the encoding selection. A cell in the matrix represents a value 1, which means the encoding supports the factor, and an empty cell represents a value −1, which means the factor is not supported.

Fig. 6.. GenoREC’s recommendation creation process:

Each component in the model identifies the most suitable output option based on the previous stage and the underlying recommendation model. Output for components (A–F) are explained in Sect. 6.1.3 (Recommendation walk through).

Fig. 7.

Utility ratings for GenoREC and Alternate stimulus across all participants and scenarios. Scenarios where GenoREC’s responses were significantly higher than the alternate stimulus, are marked with *.