Application
application/structure-qa
Quality assessment and ranking of predicted structures and interfaces.
Notes
- dockq DockQ, the reference interface-quality metric that the topological scorers in this corpus are trained and judged against. Environment
- dssp DSSP, providing the mkdssp binary for secondary-structure assignment and solvent accessibility from a structure. Environment
- mmseqs2 MMseqs2, for fast sequence clustering — the tool that makes leakage-safe benchmark splits practical. Environment
- open-topodockq-featurizer The open persistent-Laplacian featurizer that replaces TopoDockQ's bytecode-only one, built over our petls-pytorch fork. Recipe
- open-topodockq-featurizer The MIT clean-room reimplementation of TopoDockQ's interface featurizer, bit-exact against the bytecode it replaces. Environment
- open-topodockq-featurizer The MIT clean-room TopoDockQ interface featurizer, bit-exact against the bytecode-only original it replaces. Package
- open-topoqa-featurizer The clean-room interface featurizer written from the TopoQA paper, packaged as the open replacement for an unlicensed upstream. Recipe
- open-topoqa-featurizer The MIT clean-room TopoQA interface featurizer, reproduced from the paper because the upstream code carries no license. Environment
- open-topoqa-featurizer The MIT clean-room TopoQA interface featurizer, written from the paper because the published implementation carries no license. Package
- open-topoqa-scorer The clean-room ProteinGAT retrain that scores what the sibling featurizer produces, packaged as the second half of an open TopoQA vertical. Recipe
- open-topoqa-scorer The MIT clean-room ProteinGAT scorer that closes the open TopoQA vertical, with weights retrained rather than reused. Environment
- open-topoqa-scorer The MIT clean-room ProteinGAT interface-quality scorer, retrained from the paper rather than reusing an unlicensed checkpoint. Package
- score-docking-poses Rank candidate protein-complex structures by predicted DockQ-like interface quality. Mold
- topodockq The TopoDockQ scorer run from a pinned git clone; MIT-licensed but built around a Python 3.8 bytecode core with no source. Environment
- TopoDockQ A persistent-Laplacian scorer predicting DockQ interface quality to rank and filter peptide-protein docking poses. Package
- Topological Deep Learning Learn the readout instead of hand-picking it — topological structure as input to a model, or as the domain the model operates on. Method
- TopoQA A topological deep learning model ranking predicted protein-complex structures by the quality of their interface. Package
- TopoQA interface-quality replication The released TopoQA checkpoint does reproduce the paper's ranking losses — and part of the margin depends on a coordinate defect. Replication experiment