RADII: Radius-Resolved Benchmark of Nanoparticle Structures for 10 Materials

RADII is a benchmark dataset for evaluating graph generative models on nanoparticle structures across different scales. It spans 10 materials and 25 radii, designed with leakage-free splits to measure model performance from in-distribution to out-of-distribution regimes. The dataset was created by kurbanintelligencelab and last updated on May 28, 2026.

Use Cases

• Benchmarking the extrapolation capabilities of graph generative models based on the radius scaling parameter.
• Studying model failure modes on out-of-distribution nanoparticle structures.
• Training or validating models for materials discovery across a continuous range of structural sizes.

Strengths

• Designed with leakage-free splits, which likely reduces evaluation bias.
• Covers a continuous scaling parameter with 25 distinct radii for systematic analysis.
• Includes structures for 10 different materials, providing some breadth.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• The dataset description is focused on the benchmark's purpose, with limited details on the specific data structure.

Provenance

Source

kurbanintelligencelab on Hugging Face.

Collection Method

Likely generated or compiled for the benchmark described in the associated KDD '26 paper.

Freshness

Last updated 2026-05-28 19:37:05.