DiffDock-Glide: Hybrid Physics-Based and Data-Driven Molecular Docking Results

A 60.8 MB dataset containing results from the DiffDock-Glide hybrid molecular docking method. The dataset, authored by Lukas Herron and last updated in April 2026, includes evaluations on the PoseBusters and DUD-E datasets. It demonstrates a method combining a modified diffusion model with Glide's postdocking minimization pipeline.

Use Cases

• Benchmarking docking algorithm performance based on evaluations against the PoseBusters dataset.
• Assessing virtual screening enrichment based on results from the DUD-E dataset against AlphaFold2 structures.
• Comparing hybrid AI/physics-based docking methods based on the described modified generative process and confidence model replacement.
• Analyzing docking pose sampling improvements for sequences without homologues in the training set.

Strengths

• Dataset size is 60.8 MB, indicating a contained and manageable download.
• The method was evaluated on two established benchmarks: PoseBusters and DUD-E.
• Results show improved sampling of near-native poses for certain target sequences.

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 methodological bias inherent to the specific hybrid approach described.

Provenance

Source

figshare

Collection Method

Results generated by the DiffDock-Glide computational method.

Time Range

Evaluation date not specified; dataset last updated April 2026.

Freshness

Last updated 2026-04 23 12:04:34; freshness should be verified.

Geography

Global (computational).