[Paper Review] Flud: a hybrid crowd-algorithm approach for visualizing biological networks

Aditya Bharadwaj, David Gwizdala+3 more|arXiv (Cornell University)|Aug 20, 2019

Data Visualization and Analytics78 references4 citations

TL;DR

Flud is a hybrid crowd–algorithm approach that combines human intuition with algorithmic suggestions to create biologically meaningful layouts of complex signaling pathways. By using a game-with-a-purpose interface, it enables non-experts to optimize network visualizations through iterative, criterion-based improvements, significantly outperforming purely algorithmic methods on graphs with high cycle counts.

ABSTRACT

Modern experiments in many disciplines generate large quantities of network (graph) data. Researchers require aesthetic layouts of these networks that clearly convey the domain knowledge and meaning. However, the problem remains challenging due to multiple conflicting aesthetic criteria and complex domain-specific constraints. In this paper, we present a strategy for generating visualizations that can help network biologists understand the protein interactions that underlie processes that take place in the cell. Specifically, we have developed Flud, an online game with a purpose (GWAP) that allows humans with no expertise to design biologically meaningful graph layouts with the help of algorithmically generated suggestions. Further, we propose a novel hybrid approach for graph layout wherein crowdworkers and a simulated annealing algorithm build on each other's progress. To showcase the effectiveness of Flud, we recruited crowd workers on Amazon Mechanical Turk to lay out complex networks that represent signaling pathways. Our results show that the proposed hybrid approach outperforms state-of-the-art techniques for graphs with a large number of feedback loops. We also found that the algorithmically generated suggestions guided the players when they are stuck and helped them improve their score. Finally, we discuss broader implications for mixed-initiative interactions in human computation games.

Motivation & Objective

• To address the challenge of creating biologically meaningful, aesthetically pleasing layouts for complex cellular signaling networks with high feedback loop counts.
• To bridge the gap between fully automated layout algorithms and manual, expert-driven design by enabling non-expert crowdworkers to contribute meaningfully.
• To develop a mixed-initiative system that balances multiple conflicting aesthetic and domain-specific criteria in network visualization.
• To evaluate whether combining human intuition with algorithmic suggestions yields superior layouts compared to state-of-the-art automated methods.

Proposed method

• Flud is an online game with a purpose (GWAP) that presents network layout tasks as a puzzle, allowing non-expert players to iteratively adjust node positions.
• The system computes a composite layout score based on five criteria: minimizing edge crossings, reducing edge lengths, dispersing disconnected node pairs, increasing node-edge separation, and maximizing downward-pointing paths.
• Players receive real-time feedback on per-criterion scores and overall layout score, enabling informed, criterion-balanced decisions.
• The system automatically assigns priority modes (e.g., downward path optimization) in a fixed, high-to-low order to prevent players from favoring low-priority strategies.
• When players are stuck, the system provides algorithmically generated suggestions via a 'clue' feature, focusing attention on a smaller, tractable subproblem.
• A simulated annealing algorithm generates initial layout suggestions and guides the crowdworkers, creating a feedback loop where human and algorithmic progress reinforce each other.

Experimental results

Research questions

• RQ1 Can non-expert crowdworkers produce high-quality biological network visualizations when guided by algorithmic suggestions and clear aesthetic criteria?
• RQ2 Does a hybrid crowd–algorithm approach outperform purely automated layout algorithms on complex networks with many feedback loops?
• RQ3 How does the system’s automatic mode assignment influence player performance and layout quality compared to user-chosen modes?
• RQ4 To what extent do algorithmic suggestions improve player performance when they are stuck during the layout task?
• RQ5 Can the integration of a biological constraint—maximizing downward-pointing paths—improve the interpretability of signaling pathway visualizations?

Key findings

• The hybrid Flud approach significantly outperformed state-of-the-art automated layout algorithms on networks with a large number of cycles, particularly in optimizing for downward-pointing paths.
• Players who received algorithmic suggestions showed measurable improvement in their scores, especially when they were stuck, indicating that suggestions provided effective guidance.
• Automatically assigning modes in priority order led to better overall layout quality than allowing players to choose modes freely, as players tended to default to less relevant criteria.
• The system’s real-time per-criterion score feedback enabled non-expert players to make informed, balanced decisions across conflicting aesthetic goals.
• The approach demonstrated replicable and consistent improvements across three test networks with varying cycle counts, validating its robustness and scalability to complex biological networks.