Create freesolv and mnsol subsets

openfe_benchmarks/data/README.md

@@ -43,4 +43,10 @@ When adding new benchmark systems:
 4. Generate ligand network mapping files, `*network*.json` (JSON format)
 5. (If necessary) Add new charge types to `PARTIAL_CHARGE_TYPES` in `data/__init__.py`
 6. Generate `ligands_<charge_type>.sdf` files with ``openfe_benchmarks/data/data_generation/charge_molecules.py``
-6. (If present) Generate `cofactors_<charge_type>.sdf` files with ``openfe_benchmarks/data/data_generation/charge_molecules.py``
+6. (If present) Generate `cofactors_<charge_type>.sdf` files with ``openfe_benchmarks/data/data_generation/charge_molecules.py``
+7. Add `experimental_*.json` as appropriate for the calculation type. Each transformation should have:
+ - a name corresponding to the alchemical network edge
+ - dg (and uncertainty is available) with "magnitude, "unit", "pint_unit_registry", ":is_custom:".
+ - reference: DOI or the like
+ - solute_*/solvent_* where * is name, smiles, inchikey, and inchi. Note that inchi* much be generated from the 3D coordinates in the ligands.sdf. See scripts in the data_generation directory for examples (For SFE data only).
+ - canonical_smiles and inchikey. Note that inchi* much be generated from the 3D coordinates in the ligands.sdf. See scripts in the data_generation directory for examples (For BFE data only).

openfe_benchmarks/data/_benchmark_systems.py

@@ -214,6 +214,9 @@ class BenchmarkData:
     reference_data : dict[str, Path] | None
         Dictionary of available reference data files where the key is the filename 'experimental*data.json',
         and the value is a Path to a reference data file.
+    subset_data : dict[str, Path] | None
+        Dictionary of available subset files where the key is the filename stem 'subset*.json',
+        and the value is a Path to a subset JSON file defining a named subset of ligands.
     details : str
         Information available in the preparation_details.md file
     """
@@ -225,6 +228,7 @@ class BenchmarkData:
     cofactors: dict[str, Path] | None
     ligand_networks: dict[str, Path] | None
     reference_data: dict[str, Path] | None
+    subset_data: dict[str, Path] | None
     details: str
 
     def __repr__(self):
@@ -234,8 +238,9 @@ def __repr__(self):
             f"protein={self.protein.name if self.protein else 'None'}, "
             f"ligands={list(self.ligands.keys())}, "
             f"cofactors={list(self.cofactors.keys()) if self.cofactors is not None else 'None'}, "
-            f"ligand_network={list(self.ligand_networks.keys()) if self.ligand_networks is not None else 'None'}"
-            f"reference_data={list(self.reference_data.keys()) if self.reference_data is not None else 'None'})"
+            f"ligand_network={list(self.ligand_networks.keys()) if self.ligand_networks is not None else 'None'}, "
+            f"reference_data={list(self.reference_data.keys()) if self.reference_data is not None else 'None'}, "
+            f"subset_data={list(self.subset_data.keys()) if self.subset_data is not None else 'None'})"
         )
 
 
@@ -269,6 +274,7 @@ def _validate_and_load_data_system(
     cofactors = {}
     ligand_networks = {}
     reference_data = {}
+    subset_data = {}
     details = None
 
     # Track all files for validation
@@ -354,6 +360,13 @@ def _validate_and_load_data_system(
             logger.debug(f"Found reference data: {filename}")
             continue
 
+        # check for subset data file (subset*.json)
+        if filename.startswith("subset") and filename.endswith(".json"):
+            subset_data[file_path.stem] = file_path
+            categorized_files.add(file_path)
+            logger.debug(f"Found subset: {filename}")
+            continue
+
     # Check for uncategorized files
     uncategorized = set(all_files) - categorized_files
     for file_path in uncategorized:
@@ -381,7 +394,7 @@ def _validate_and_load_data_system(
             raise ValueError(
                 f"Uncategorized JSON file '{filename}' found in system '{system_name}' "
                 f"in benchmark set '{benchmark_set}'. Expected format: "
-                f"'*network*.json' for ligand networks."
+                f"'*network*.json' for ligand networks, 'experimental*data.json', or 'subset*.json'."
             )
 
         raise ValueError(
@@ -425,6 +438,7 @@ def _validate_and_load_data_system(
         cofactors=cofactors or None,
         ligand_networks=ligand_networks or None,
         reference_data=reference_data or None,
+        subset_data=subset_data or None,
         details=details,
     )
 

openfe_benchmarks/data/benchmark_systems/solvation_set/freesolv/PREPARATION_DETAILS.md

@@ -1,17 +1,23 @@
-Done by Joshua Horton on 2026-02-12
+Done by Joshua Horton on 2026-02-12 (subset regenerated 2026-04-02)
 
 All ligands extracted from the FreeSolv database version https://github.com/MobleyLab/FreeSolv/releases/tag/v0.52
 
-
-
-## Notes
+## Partial Charges
 
 The reference data was generated using the [generate_freesolv_exp_data.py](../../../data_generation/generate_freesolv_exp_data.py) script using the [conda-lock_linux-64.yml](../../../data_generation/conda-lock_linux-64.yml) environment. 
 Charges were generated using the [charge_freesolv.py](../../../data_generation/charge_freesolv.py) script using the [conda-lock_linux-64.yml](../../../data_generation/conda-lock_linux-64.yml) environment. 
 Some ligands could not be charged with all methods, the following lists the ligands that could not be charged with each method:
 
-- am1bccelf10_oe:
-    - mobley_7176248
+Charging failures:
+- `am1bccelf10_oe`: mobley_7176248
+- `am1bcc_at`: mobley_9741965
+
+## Subsets
+
+Regenerate all subsets: run `python define_freesolv_mnsol_openff_subsets.py` in `data_generation/`.
+
+### subset_openff_filtered.json
+588 neutral solutes retained from the full FreeSolv v0.52 database, all measured in water. Excluded 54 entries: 47 with elements outside the OpenFF chemical space, 5 with disqualifying SMIRKS patterns, and 2 with undefined stereochemistry. Each entry is a unique solute with a single experimental aqueous hydration free energy measurement.
 
-- am1bcc_at:
-    - mobley_9741965
+### subset_openff_small.json
+188 unique solutes drawn from `subset_openff_filtered`. Selection seeded the 188 solutes present in both the FreeSolv and MNSol filtered pools (by OpenFF SMILES isomorphism).