EviCYP: 10,996 Molecular Samples for Cytochrome P450 Substrate Prediction

10,996 samples, including 4388 substrates and 2880 nonsubstrates curated from 1629 publications, plus 3728 pseudonegative samples, for nine major CYP isoforms. The dataset was created by Yingjie Yang and published on figshare in March 2026. It supports the EviCYP framework, which integrates evidential deep learning and vector quantization to predict substrates with uncertainty estimates.

Use Cases

• Train models for CYP substrate classification based on multimodal molecular representations and enzyme sequences.
• Benchmark uncertainty quantification methods in computational biology based on the evidential learning framework.
• Study the metabolism profiles of drug candidates across nine major CYP isoforms.
• Investigate feature compression and redundancy reduction techniques using the described vector quantization approach.

Strengths

• Dataset size of 10,996 samples, including curated positive and negative examples.
• Data sourced from 1629 publications, suggesting a literature-based compilation.
• Associated model achieved an average AUROC of 0.9500 on an internal test set.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect publication bias inherent to the source literature.

Provenance

Source

figshare, author Yingjie Yang.

Collection Method

Curated from 1629 publications, supplemented with pseudonegative samples.

Freshness

Last updated 2026-03-17 19:21:17; freshness should be verified.