BeeEcoTox: Ecotoxicity Predictions for 1,139 Agrochemicals to Bees

A dataset of 1,139 agrochemicals used to train a multimodal graph-learning framework for predicting ecotoxicity to bees. The model, developed by Xuanlin Chen and deployed in 2026, fuses semantic features from ChemFM with molecular graphs and structural features. It addresses class imbalance via cost-sensitive learning and achieves an area under the curve of 0.91 and a recall of 0.90.

Use Cases

• Predicting agrochemical ecotoxicity to bees based on fused molecular graph and structural features.
• Training cost-sensitive learning models based on the inherent class imbalance in the curated dataset.
• Identifying key toxicophores for model explainability using the GNNExplainer approach mentioned.
• Benchmarking new ecotoxicity prediction models against the reported state-of-the-art performance metrics.

Strengths

• Contains structural features from 1,139 agrochemicals.
• Model performance is reported with specific metrics: area under the curve of 0.91 and recall of 0.90.
• Addresses class imbalance through a documented cost-sensitive learning approach.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect bias inherent to the specific curation process for the 1,139 agrochemicals.

Provenance

Source

Xuanlin Chen via figshare.

Collection Method

Curated dataset used to train the BeeEcoTox framework; specific collection method not detailed.

Time Range

Temporal coverage of the agrochemical data is not specified.

Freshness

Last updated 2026-04-15 06:43:27; freshness should be verified.

Geography

Spatial coverage is not specified.