SC-DTI: Processed Drug-Target Interaction Benchmarks for Machine Learning

1.1 GB of processed data files support the study "SC-DTI: A Structure–Context Multimodal Framework with Local–Global Collaborative Modeling for Drug–Target Interaction Prediction". The dataset contains processed records and features from three public benchmarks: DrugBank, BioSNAP, and Davis. It was authored by Tianqi Chen and last updated on 2026-05-30.

Use Cases

• Train binary classification models for drug-target interaction based on processed interaction records.
• Evaluate multimodal machine learning frameworks using precomputed ligand and protein features.
• Benchmark prediction algorithms against established datasets like DrugBank, BioSNAP, and Davis.

Strengths

• Includes processed data from three established public benchmark datasets: DrugBank, BioSNAP, and Davis.
• Data is split into training (80%), validation (within training), and test (20%) sets for direct model evaluation.
• Files are provided in multiple formats (MD, PKL, TXT) totaling 1.1 GB.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Processed from three public benchmark datasets: DrugBank, BioSNAP, and Davis.

Collection Method

Original datasets were obtained from public sources and processed following the study's protocol, including random negative sample generation for DrugBank.

Freshness

Last updated 2026-05-30 12:17:43; freshness should be verified.