When exploring data interactively, an analyst performs a continuous sequence of operations: zooming into a specific genomic region, filtering out low-expression genes, adjusting a 3D slicing plane, or rearranging network nodes. However, traditional methods of sharing these discoveries are lossy. A static screenshot discards interactivity entirely, while sharing raw data forces the recipient to manually recreate the exact steps to find the insight. As noted by Tom Mansion et al. in Meet DebiAI: A Versatile Open-Source Tool for Streamlined Data Analysis, Visualization, and ML Model Evaluation, web-based platforms frequently struggle to maintain user configurations across sessions, disrupting analytical workflows and breaking continuity.
To bridge this gap, a prominent research direction focuses on declarative state serialization. Instead of capturing the visual output (pixels), these systems capture the underlying interaction state as a structured, portable data object. This approach treats a visualization session not as a static image, but as a reproducible, executable environment.
The Core Mechanism: Capturing the State Vector
The fundamental intuition behind state serialization is that an interactive visualization can be modeled as a deterministic function of two inputs: the underlying dataset ($D$) and the current configuration state vector ($S$).
$$\text{Visualization} = f(D, S)$$
If a system can cleanly isolate and serialize $S$, it can perfectly reconstruct the exact visual and analytical environment anywhere, provided the recipient has access to $D$.
[User Interactions]
│ (Zoom, Filter, Pan, Select)
▼
┌────────────────────────────────────────┐
│ Interactive Application State │
└────────────────────────────────────────┘
│
│ Serialize to Declarative Format
▼
┌────────────────────────────────────────┐
│ JSON / XML / URL State Object │ ──► [Share / Version Control]
└────────────────────────────────────────┘
│
│ Deserialize & Rehydrate
▼
┌────────────────────────────────────────┐
│ Identical Interactive Session │
└────────────────────────────────────────┘
In modern web architectures, this state vector is typically represented as a declarative JSON object. For example, Huamin Qu et al. demonstrate in PyGWalker: On-the-fly Assistant for Exploratory Visual Data Analysis how interactive dashboard states can be saved as portable JSON configurations. This mechanism allows fluid, drag-and-drop graphical user interface (GUI) interactions to be preserved in a format compatible with version control and programmatic environments like computational notebooks.
Similarly, Peter Kerpedjiev et al. apply this to multi-scale genomic data in HiGlass: web-based visual exploration and analysis of genome interaction maps. By storing data sources, track arrangements, and view synchronization rules in a structured JSON object, they enable researchers to transmit complex, linked visualization sessions rather than context-free images.
This architecture is highly adaptable and scales across different dimensions of visualization:
- 3D Graphics Pipelines: Prabhu Ramachandran highlights in Mayavi: Making 3D Data Visualization Reusable that robust persistence mechanisms are required to reliably capture and restore the complete state of multi-layered 3D pipelines across different sessions.
- Augmented Reality (AR): In Multi-User Mobile Augmented Reality for Cardiovascular Surgical Planning, Pratham Darrpan Mehta et al. serialize 3D model orientations, scales, and active slicing planes. This ensures that surgical collaborators can consistently return to or share specific spatial configurations of a virtual organ.
From Snapshots to Asynchronous Collaboration
When the state vector is serialized into a compact string, it can be embedded directly into a URL or stored in a lightweight cloud database. This capability transforms solitary data exploration into an asynchronous, collaborative process.
A pioneering implementation of this concept was introduced by Frank van Ham et al. in ManyEyes: a Site for Visualization at Internet Scale. By capturing the complete interactive state—including zoom levels, filters, and selections—within a unique URL, the platform allowed users to share specific visual insights as persistent snapshots that others could revisit, discuss, and build upon in context.
More recently, Siwei Li et al. applied this pattern to large-scale network graphs in Argo Lite: Open-Source Interactive Graph Exploration and Visualization in Browsers. By serializing node positions, visual attributes, and visibility settings into cloud-stored state objects accessible via unique URLs, they achieved instantaneous, Google-Doc-style collaboration for graph exploration.
Beyond simple sharing, state persistence provides workspace continuity across time and hardware boundaries:
- Session Continuity: 講演要約 長塚 隆, in describing the G-Portal experience in The use of digital libraries in learning: The G-Portal experience, emphasizes that preserving the exact state of a user's workspace (such as map zoom levels paired with specific metadata filters) is crucial for maintaining the continuity of long-term research tasks.
- Cross-Device Migration: Soo-Yeon Ji et al. demonstrate in Designing a collaborative visual analytics system to support users’ continuous analytical processes that capturing comprehensive snapshots of system parameters, active filters, and user annotations allows an analytical session to be reconstructed and transferred seamlessly across entirely different hardware environments, such as migrating from a solitary desktop to a collaborative multi-touch table.
Binding Visualization to Computational Provenance
In scientific domains, visual exploration does not happen in a vacuum; it is deeply intertwined with underlying computational pipelines. True reproducibility requires that the serialized state captures not just the UI state, but the entire analytical provenance.
To achieve this, researchers have developed frameworks that bind visual configurations with data transformations and model parameters. In Epiviz: a view inside the design of an integrated visual analysis software for genomics, Florin Chelaru et al. design a system where saved analysis states encompass visualizations, code customizations, and data transformations simultaneously. This holistic snapshot ensures that complex genomic analyses can be replicated and verified by collaborators.
A similar approach is taken in transcriptomics by Qin Zhu et al. in PIVOT: platform for interactive analysis and visualization of transcriptomics data, where the entire state of an analysis—including all steps, algorithms, and parameters—is saved as a single object to enable automated report generation and reliable sharing.
This pattern of unifying workflow logic and state serialization appears across diverse scientific disciplines:
- Gene Ontology: Ziyin Xin et al. show in MonaGO: a novel gene ontology enrichment analysis visualisation system that serializing the entire interactive state into a portable format allows specific exploratory states to be shared or revisited without re-running the underlying, computationally expensive analysis.
- Spatial Modeling: In GeoExplainer: A Visual Analytics Framework for Spatial Modeling Contextualization and Report Generation, Fan Lei et al. combine model configurations, explored information, and generated reports into a single reloadable state, ensuring that the narrative explanations provided to researchers remain fully auditable.
- Molecular Dynamics: Gergely Varga et al. demonstrate in Web-based Metaprogrammable Frontend for Molecular Dynamics Simulations that storing all aspects of a simulation setup, including parameters, tool versions, and workflow logic, in a structured database guarantees the reproducibility of computational research.
- Affective Sensing: Even in multimodal systems, Gordon McIntyre et al. show in A composite framework for affective sensing how XML serialization can be leveraged to persist and restore the state of complex projects and their constituent software components, facilitating iterative development.
By treating the interactive interface as a state machine whose configuration can be externalized, serialized, and rehydrated, these systems successfully transform ephemeral user interactions into permanent, shareable, and reproducible scientific assets.
Key Foundational Work
The use of digital libraries in learning: The G-Portal experience (2005) — contributed 2 concept elements to this direction.
ManyEyes: a Site for Visualization at Internet Scale (2007) — contributed 1 concept elements to this direction.