Your message has been sent.
We'll get back to you as soon as possible.
Terminology
Two terms that sound similar but describe different things.
File format
GraphML stands for Graph Markup Language. It is an XML-based format for describing graph structures: nodes, edges, labels, and application-specific data attributes. When graph data needs to be exchanged, stored, or inspected across tools, GraphML provides a portable, schema-validated representation.
Spec: graphml.graphdrawing.org
Machine learning
Graph ML refers to graph machine learning — a family of methods that learn from graph-structured data. A model can learn from the attributes of individual nodes and edges, but also from the topology of the graph itself: the connections, neighborhoods, and paths between entities. Applications include citation networks, molecules, knowledge graphs, fraud detection, recommenders, and supply chains.
Similar names, but the two meet in practice: graph machine learning starts with graph data, and that data often needs to be visualized, exchanged, debugged, and explained.
Explainer
Graph ML explained through examples rather than equations. A graph represents entities and relationships. A model can learn not only from attributes of individual items, but also from how those items are connected.
Predict the class or label of individual nodes — a paper's topic in a citation network, a user's role in a social graph, or a transaction's risk in a financial network.
Estimate whether an edge should exist between two nodes — useful for recommender systems, knowledge graph completion, fraud ring detection, and protein interaction prediction.
Classify or score an entire graph — determine whether a molecule is soluble, whether a process diagram has anomalies, or which software dependency graph has the highest risk profile.
Why it matters
Graph machine learning workflows involve adjacency matrices, tensors, embeddings, and model outputs — essential for training, but they can obscure the structure that domain experts care about.
| node_id | feat_vec | pred | score |
|---|---|---|---|
| n0 | [0.12, 0.88…] | A | 0.91 |
| n1 | [0.09, 0.76…] | A | 0.83 |
| n2 | [0.44, 0.31…] | B | 0.67 |
| n3 | [0.71, 0.14…] | C | 0.59 |
| n4 | [0.33, 0.50…] | B | 0.78 |
Structure and neighborhoods are hidden.
Color = predicted class
With yFiles, these views become part of an app: users can pan, zoom, select nodes, inspect features, run layouts, filter large networks, and connect visual interactions to model results.
Workflow
GraphML is one practical interchange and inspection format — not the only one, but a well-specified option when you need XML-based portability.
CSV, JSON, RDF, NetworkX, PyG, DGL, or domain databases
Nodes, edges, labels, weights, feature metadata
XML-based portable interchange for inspection and tooling
yFiles: layout, interaction, pan, zoom, filter
Predictions, confidence, embeddings, explanations
Round-trip: from yFiles, users can annotate, re-export as GraphML, and feed corrections back into the ML pipeline.
Interactive demo
Below, you will find a demo that shows how Graph Machine Learning can be used to identify patterns in a graph structure. yFiles can use the classification to apply a matching layout, further highlighting the identified patterns.
The demo uses a machine learning model and trains it inside the browser to recognize patterns in a graph structure. yFiles then uses the classification to apply either a tree-, force-directed, or orthogonal layout.
Spot the unclassified structures in gray? Click Infer Colors in the top right of the demo to classify them with the model. Then experiment by modifying the graph model and rerunning the classifier and layout to see your changes in action.
More examples
Each scenario demonstrates a different graph ML task and a different reason why keeping the graph visible matters.
Link prediction
Select two nodes to see predicted link probability. Candidate edges appear as dashed lines. Useful for recommender systems, knowledge graph completion, and fraud ring investigations.
Graph classification
Atoms as nodes, bonds as edges, whole-graph prediction. Useful for drug discovery, toxicity screening, and material property estimation. Color encodes atom type; prediction shown at graph level.
Representation learning
Two linked views: graph structure and 2D embedding projection. Selecting a node in either view highlights it in the other. Shows how graph ML transforms structural context into learned representations.
Explainability
A prediction is shown alongside a highlighted explanatory subgraph — the k-hop neighborhood or attention-weighted edges most influential in reaching the result. Positions yFiles as an inspection and explanation layer.
yFiles
yFiles is family of graph visualization libraries for building interactive graph applications for the web, mobile, and the desktop. For graph machine learning projects, it provides the visualization layer: layout, interaction, styling, editing, and integration.
Built-in GraphML I/O: load graph structure and data attributes, save annotated results.
Map class labels, confidence scores, embeddings, and explanations directly to node and edge styles, heat maps, and more.
Organic, hierarchical, orthogonal, circular, and radial layouts that make complex structures readable at a glance.
Encode uncertainty, confidence bands, or attention weights as visual attributes with full style control.
Pan, zoom, click-to-inspect, neighborhood highlighting, and filterable large networks.
Works with desktop and cross platform technologies, as well as web technologies like React, Angular, Vue, or plain JavaScript. Embed graph views in research tools, dashboards, and web applications.
For researchers
Graph ML visualization for academic and technical users means preserving access to data, not hiding it behind decorative visuals.
Applications
01
Inspect graph datasets, labels, model outputs, and explanations during model development.
02
Find mislabeled nodes, suspicious neighborhoods, data leakage, or unexpected cluster formation.
03
Explore entity relationships, inferred links, and confidence scores across large knowledge bases.
04
Investigate suspicious subgraphs, connected accounts, and transaction patterns in financial networks.
05
Visualize molecular graphs and predicted graph-level properties for chemistry and biology teams.
06
Explain why users, items, papers, or products are connected in bipartite or knowledge graphs.
Further reading
Specification, introductory material, and yFiles demos for working with GraphML and graph machine learning in practice.
Yes. yFiles for HTML includes built-in GraphML I/O support. You can load a GraphML file, attach custom data attributes, and save the annotated graph back to GraphML. The GraphML demo shows this in the browser.
GraphML is a portable XML format for graph data. Graph machine learning uses graph structure to make predictions about nodes, links, and entire graphs.
This guide explains both — and shows how yFiles can help you visualize, inspect, and build interactive graph machine learning applications.
Connect with our Customer Success Team regarding your ideas or projects.
Try yFiles free of charge.
