DR-MHEA-GCN: Graph Data for Chemical Bond Classification in 2D Nanomaterials

A graph dataset of approximately 3,000 two-dimensional nanomaterials, encoded as atomistic graphs with atomic and bond descriptors. Developed by Liying Cui and last updated in May 2026, it supports a machine learning model achieving 95.3% test accuracy for chemical bond identification.

Use Cases

• Training graph neural networks for chemical bond classification based on atomic and bond descriptors.
• Benchmarking new models against a reported 95.3% test accuracy for bond identification in 2D systems.
• Analyzing the role of atomic electronegativity in bond prediction, as highlighted by the model's interpretable analysis.
• Accelerating materials analysis by approximately 3 orders of magnitude over conventional DFT-based methods using the provided graph representations.

Strengths

• Model performance is quantified with a 95.3% test accuracy and 98.1% agreement on extrapolative data.
• The dataset is derived from approximately 3,000 distinct 2D nanomaterial structures.
• The underlying method accelerates analysis by approximately 3 orders of magnitude over conventional DFT-based methods.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset's 47.2 KB size suggests a limited scope, likely containing model outputs or descriptors rather than raw atomic coordinates.

Provenance

Source

figshare

Collection Method

Encoded from approximately 3,000 2D nanomaterials using physically motivated atomic and bond descriptors.

Freshness

Last updated 2026-05-18 19:14:21; freshness should be verified.