posebench-fast 0.1.1__py3-none-any.whl → 0.1.3__py3-none-any.whl

posebench_fast/__init__.py

@@ -29,7 +29,7 @@ from posebench_fast.metrics.rmsd import (
     time_limit,
 )
 
-__version__ = "0.1.1"
+__version__ = "0.1.3"
 
 __all__ = [
     # RMSD

posebench_fast/metrics/rmsd.py

@@ -1,4 +1,4 @@
-# Taken from https://github.com/RMeli/spyrmsd and https://github.com/gcorso/DiffDock/
+# Based on https://github.com/RMeli/spyrmsd
 
 import signal
 from contextlib import contextmanager

posebench_fast-0.1.3.dist-info/METADATA

@@ -0,0 +1,71 @@
+Metadata-Version: 2.4
+Name: posebench-fast
+Version: 0.1.3
+Summary: Fast docking evaluation metrics: symmetry-corrected RMSD and PoseBusters filters
+Keywords: docking,rmsd,posebusters,molecular-docking,benchmark
+Author: Nikolenko
+License-Expression: MIT
+Requires-Dist: numpy>=1.21
+Requires-Dist: pandas>=1.3
+Requires-Dist: spyrmsd>=0.6
+Requires-Dist: rdkit>=2022.3
+Requires-Dist: torch>=1.10
+Requires-Dist: posebusters>=0.2
+Requires-Dist: tqdm>=4.60
+Requires-Dist: pytest>=7.0 ; extra == 'dev'
+Requires-Dist: pytest-cov>=4.0 ; extra == 'dev'
+Requires-Dist: ruff>=0.4 ; extra == 'dev'
+Requires-Python: >=3.11
+Provides-Extra: dev
+Description-Content-Type: text/markdown
+
+# posebench-fast
+
+[![PyPI version](https://badge.fury.io/py/posebench-fast.svg)](https://pypi.org/project/posebench-fast/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
+Fast docking evaluation metrics: symmetry-corrected RMSD and lightweight PoseBusters filters.
+
+## Installation
+
+```bash
+pip install posebench-fast
+```
+
+## Features
+
+- **Symmetry-corrected RMSD** — accounts for molecular symmetry (benzene, carboxylates, etc.)
+- **Fast PoseBusters filters** — 4 key checks in ~10ms instead of full 27-test suite (~1-2s)
+
+## Usage
+
+```python
+from posebench_fast import compute_all_isomorphisms, get_symmetry_rmsd_with_isomorphisms
+
+# Symmetry-corrected RMSD
+isomorphisms = compute_all_isomorphisms(rdkit_mol)
+rmsd = get_symmetry_rmsd_with_isomorphisms(true_coords, pred_coords, isomorphisms)
+```
+
+```python
+from posebench_fast import check_intermolecular_distance
+
+# Fast filters: not_too_far_away, no_clashes, no_volume_clash, no_internal_clash
+results = check_intermolecular_distance(
+    mol_orig=rdkit_mol,
+    pos_pred=pred_positions,      # (n_samples, n_atoms, 3)
+    pos_cond=protein_positions,   # (n_protein_atoms, 3)
+    atom_names_pred=lig_atoms,
+    atom_names_cond=prot_atoms,
+)
+```
+
+## Related
+
+- [PoseBench](https://github.com/BioinfoMachineLearning/PoseBench) — full benchmark suite
+- [PoseBusters](https://github.com/maabuu/posebusters) — full 27-test validation
+- [spyrmsd](https://github.com/RMeli/spyrmsd) — symmetry RMSD algorithms
+
+## License
+
+MIT

{posebench_fast-0.1.1.dist-info → posebench_fast-0.1.3.dist-info}/RECORD

@@ -1,11 +1,11 @@
-posebench_fast/__init__.py,sha256=_ExeWnJgtI_-Op6Xf07Jgr1gi6WaissAOnrCoCgFs6w,1264
+posebench_fast/__init__.py,sha256=CFNN5MzTlkjRbio39aU4bSj6lpYHPs3gd-Baz9woXms,1264
 posebench_fast/datasets/__init__.py,sha256=-29X5OLrM8eE-bXnB-S_RC4ROjZq0d0lluVAOKtHkRM,44
 posebench_fast/filters/__init__.py,sha256=Qm9CK25oMyWwV0gopk7qcU0YQw8cnx9qJFJFQjybSF8,337
 posebench_fast/filters/fast_filters.py,sha256=qfMNIAyCDXtVh5oDDl-iC-1WKxJmzN55LVJTwbolHE4,17428
 posebench_fast/metrics/__init__.py,sha256=g49mQpR1uSy6v3gFtarTTVcDKeQXlB3ZONCntFLlRs4,765
 posebench_fast/metrics/aggregation.py,sha256=QZNRlp40kpdcxjC5H9WyR6KWFzboOHcmcCke8AEdD-w,13046
-posebench_fast/metrics/rmsd.py,sha256=BmHPzUCBQCEAvG8slbs7mXASZV8_ROn7ljplLcH0OFw,7410
+posebench_fast/metrics/rmsd.py,sha256=LwQeSCHtIvXQswnUqAhXWw5-5_43M61k5aQgUhnFoOo,7368
 posebench_fast/utils/__init__.py,sha256=TMMehJTihRawWfPW_qBEQIVzpf-Z92QujVAcgNydylI,44
-posebench_fast-0.1.1.dist-info/WHEEL,sha256=5DEXXimM34_d4Gx1AuF9ysMr1_maoEtGKjaILM3s4w4,80
-posebench_fast-0.1.1.dist-info/METADATA,sha256=_Lmmfdd_F9wUsgLV7jzTDGVnOO2AhcdeNxfvFuaYf2M,7374
-posebench_fast-0.1.1.dist-info/RECORD,,
+posebench_fast-0.1.3.dist-info/WHEEL,sha256=5DEXXimM34_d4Gx1AuF9ysMr1_maoEtGKjaILM3s4w4,80
+posebench_fast-0.1.3.dist-info/METADATA,sha256=OLzzq9oHe504c4X83xE7JaIQGbkHiy5yzmHVVBP_KSU,2212
+posebench_fast-0.1.3.dist-info/RECORD,,

posebench_fast-0.1.1.dist-info/METADATA

@@ -1,201 +0,0 @@
-Metadata-Version: 2.4
-Name: posebench-fast
-Version: 0.1.1
-Summary: Fast docking evaluation metrics: symmetry-corrected RMSD and PoseBusters filters
-Keywords: docking,rmsd,posebusters,molecular-docking,benchmark
-Author: Nikolenko
-License-Expression: MIT
-Requires-Dist: numpy>=1.21
-Requires-Dist: pandas>=1.3
-Requires-Dist: spyrmsd>=0.6
-Requires-Dist: rdkit>=2022.3
-Requires-Dist: torch>=1.10
-Requires-Dist: posebusters>=0.2
-Requires-Dist: tqdm>=4.60
-Requires-Dist: pytest>=7.0 ; extra == 'dev'
-Requires-Dist: pytest-cov>=4.0 ; extra == 'dev'
-Requires-Dist: ruff>=0.4 ; extra == 'dev'
-Requires-Python: >=3.11
-Provides-Extra: dev
-Description-Content-Type: text/markdown
-
-# posebench-fast
-
-Fast and accurate docking evaluation metrics for molecular docking benchmarks.
-
-[![PyPI version](https://badge.fury.io/py/posebench-fast.svg)](https://pypi.org/project/posebench-fast/)
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
-
-## Why posebench-fast?
-
-This is a **lightweight alternative** to the original [PoseBench](https://github.com/BioinfoMachineLearning/PoseBench) benchmark suite, focused on speed and practical metrics.
-
-### posebench-fast vs PoseBench
-
-| Feature | PoseBench (original) | posebench-fast |
-|---------|---------------------|----------------|
-| **Purpose** | Full benchmark suite with datasets | Evaluation metrics only |
-| **Dependencies** | Heavy (DGL, ESM, etc.) | Minimal (rdkit, torch) |
-| **Installation** | Complex setup | `pip install posebench-fast` |
-| **Speed** | Full pipeline | 100x faster filters |
-| **Use case** | Reproduce paper results | Integrate into your pipeline |
-
-**When to use PoseBench:** Reproducing benchmark results, running full evaluation suite
-
-**When to use posebench-fast:** Quick pose filtering, custom pipelines, production inference
-
----
-
-Standard RMSD calculations can **severely penalize correct predictions** for symmetric molecules. Consider a benzene ring or a carboxylate group - rotating them by 180° gives chemically identical structures, but naive RMSD treats them as completely different poses.
-
-### The Symmetry Problem
-
-```
-     O              O
-     ‖              ‖
- R—C—O⁻   vs   ⁻O—C—R
-
- Same molecule, different atom ordering → High RMSD with naive calculation!
-```
-
-**posebench-fast** solves this by:
-1. Computing all valid graph isomorphisms of the molecule
-2. Finding the atom permutation that minimizes RMSD
-3. Returning the **true** structural difference
-
-### Fast vs Full PoseBusters
-
-Full [PoseBusters](https://github.com/maabuu/posebusters) validation runs 27 tests including expensive energy calculations. For rapid screening of thousands of docking poses, this is too slow.
-
-**posebench-fast** provides 4 key physics-based filters that catch most invalid poses in milliseconds:
-
-| Filter | What it catches |
-|--------|-----------------|
-| `not_too_far_away` | Ligand flew away from binding site (distance > 5Å) |
-| `no_clashes` | Atomic clashes with protein (overlapping VDW radii) |
-| `no_volume_clash` | Volume overlap > 5% with protein |
-| `no_internal_clash` | Invalid bond lengths/angles within ligand |
-
-**Speed comparison:**
-- Full PoseBusters: ~1-2 sec/pose
-- posebench-fast: ~5-10 ms/pose (100x faster)
-
-## Installation
-
-```bash
-pip install posebench-fast
-```
-
-Or with uv:
-```bash
-uv add posebench-fast
-```
-
-## Usage
-
-### Symmetry-Corrected RMSD
-
-```python
-from posebench_fast import compute_all_isomorphisms, get_symmetry_rmsd_with_isomorphisms
-
-# Compute isomorphisms once per molecule (cache this!)
-isomorphisms = compute_all_isomorphisms(rdkit_mol)
-
-# Compute symmetry-corrected RMSD for any number of predictions
-rmsd = get_symmetry_rmsd_with_isomorphisms(true_coords, pred_coords, isomorphisms)
-
-# Example: benzene with 12 equivalent atom orderings
-# Naive RMSD: 2.4 Å (WRONG - penalizes valid rotation)
-# SymmRMSD:   0.3 Å (CORRECT - finds best match)
-```
-
-### Fast PoseBusters Filters
-
-```python
-from posebench_fast import check_intermolecular_distance
-
-results = check_intermolecular_distance(
-    mol_orig=rdkit_mol,
-    pos_pred=predicted_positions,    # (n_samples, n_atoms, 3)
-    pos_cond=protein_positions,      # (n_protein_atoms, 3)
-    atom_names_pred=ligand_atoms,    # ['C', 'N', 'O', ...]
-    atom_names_cond=protein_atoms,   # ['C', 'CA', 'N', ...]
-)
-
-# Results dict contains:
-# - not_too_far_away: [True, True, False, ...]  # ligand near protein?
-# - no_clashes: [True, False, True, ...]        # no atomic overlaps?
-# - no_volume_clash: [True, True, True, ...]    # no volume overlap?
-# - no_internal_clash: [True, True, False, ...] # valid geometry?
-# - is_buried_fraction: [0.8, 0.6, 0.2, ...]    # how buried is ligand?
-```
-
-### Batch Metrics with Filtering
-
-```python
-from posebench_fast import get_final_results_for_df, filter_results_by_fast
-
-# Filter predictions to keep only physically valid ones
-filtered = filter_results_by_fast(predictions)
-
-# Compute benchmark metrics
-metrics_df, scored = get_final_results_for_df(
-    predictions,
-    score_names=['error_estimate_0'],  # ranking score
-    posebusters_filter=True,
-    fast_filter=True
-)
-
-# metrics_df contains:
-# | ranking          | RMSD<2Å | RMSD<5Å | SymRMSD<2Å | avg RMSD | ...
-# |------------------|---------|---------|------------|----------|
-# | error_estimate_0 | 0.45    | 0.78    | 0.52       | 3.2      |
-# | ..._fast         | 0.48    | 0.82    | 0.58       | 2.9      |  <- after filtering
-```
-
-## API Reference
-
-### RMSD Functions
-
-| Function | Description |
-|----------|-------------|
-| `compute_all_isomorphisms(mol)` | Get all graph isomorphisms for symmetry correction |
-| `get_symmetry_rmsd_with_isomorphisms(coords1, coords2, iso)` | Fast symmetry-corrected RMSD |
-| `get_symmetry_rmsd(mol, coords1, coords2)` | Convenience wrapper (computes isomorphisms internally) |
-
-### Filter Functions
-
-| Function | Description |
-|----------|-------------|
-| `check_intermolecular_distance(...)` | All 4 fast filters + buried fraction |
-| `check_volume_overlap(...)` | Volume overlap only |
-| `check_geometry(...)` | Internal geometry validation only |
-
-### Aggregation Functions
-
-| Function | Description |
-|----------|-------------|
-| `get_final_results_for_df(...)` | Compute full metrics table |
-| `filter_results_by_fast(...)` | Keep best poses by fast filter count |
-| `filter_results_by_posebusters(...)` | Keep best poses by full PB count |
-| `get_best_results_by_score(...)` | Select best pose per molecule by score |
-
-## When to Use What
-
-| Scenario | Recommendation |
-|----------|----------------|
-| Quick screening of 1000s of poses | `check_intermolecular_distance` |
-| Final benchmark reporting | Full PoseBusters + `get_final_results_for_df` |
-| Symmetric molecules (rings, COO⁻, etc.) | Always use `get_symmetry_rmsd_*` |
-| Speed-critical pipelines | `filter_results_by_fast` |
-
-## Related Projects
-
-- **[PoseBench](https://github.com/BioinfoMachineLearning/PoseBench)** - Original comprehensive benchmark suite for molecular docking. Use if you need full datasets and reproducible benchmark runs.
-- [spyrmsd](https://github.com/RMeli/spyrmsd) - Symmetry-corrected RMSD algorithms
-- [PoseBusters](https://github.com/maabuu/posebusters) - Full physical validity validation (27 tests)
-- [DiffDock](https://github.com/gcorso/DiffDock) - Fast filter implementations origin
-
-## License
-
-MIT