Fix scoring bug, and tests (#65)

* test: make data loader test run and add split ids to index

* test: make plot use old split file

* test: update index length

* drop atom3d

* doc: column name

* refactor: drop unused files

* test: drop empty file

* test: add dataclass tests from pinder

* test: add transform tests

* log test error

* fix plot

* add torch dependency

* add torch to data requirements

* lint

* FIX: bug in scoring

* FIX: Bug in scoring

* fix: don't allow rigid docking on holo (just to be safe)

* type

* chore: fix hyperlink

* fix: fix failing tests

* fix: linting

* fix: mount data discrepancies

* fix: load split in index to avoid global issue

* lint and type

---------

Co-authored-by: OleinikovasV <[email protected]>
Co-authored-by: yusuf1759 <[email protected]>

docs/tutorial/api.ipynb

@@ -161,7 +161,7 @@
     "of a column name with some given value.\n",
     "Only those rows, that fulfill all conditions, are returned.\n",
     "See the description of\n",
-    "[`pandas.read_parquet()`]https://pandas.pydata.org/docs/reference/api/pandas.read_parquet.html\n",
+    "[`pandas.read_parquet()`](https://pandas.pydata.org/docs/reference/api/pandas.read_parquet.html)\n",
     "for more information on the filter syntax."
    ]
   },
@@ -285,7 +285,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -402,7 +402,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.10.14"
+   "version": "3.10.15"
   }
  },
  "nbformat": 4,

pyproject.toml

@@ -66,7 +66,6 @@ dev = [
 ]
 loader = [
     "torch",
-    "atom3d",
 ]
 plots = [
     "matplotlib",

requirements_data.txt

@@ -2,3 +2,4 @@
   tabulate
   pdb-validation @ git+https://git.scicore.unibas.ch/schwede/ligand-validation.git
   mmpdb @ git+https://github.com/rdkit/mmpdb.git
+  torch

src/plinder/core/loader/dataset.py

@@ -14,6 +14,9 @@
 from plinder.core.scores import query_index
 from plinder.core.scores.query import FILTERS
 from plinder.core.structure.structure import Structure
+from plinder.core.utils.log import setup_logger
+
+LOG = setup_logger(__name__)
 
 
 class PlinderDataset(Dataset):  # type: ignore
@@ -44,9 +47,9 @@ def __init__(
         ] = structure_featurizer,
         **kwargs: Any,
     ):
-        self._system_ids = list(
-            set(query_index(splits=[split], filters=filters)["system_id"])
-        )
+        index = query_index(splits=[split], filters=filters)
+        LOG.info(f"Loading {index.system_id.nunique()} systems")
+        self._system_ids = list(set(index["system_id"]))
         self._num_examples = len(self._system_ids)
 
         self._featurizer = featurizer

src/plinder/core/scores/index.py

@@ -6,7 +6,6 @@
 from duckdb import sql
 
 from plinder.core.scores.query import FILTERS, make_query
-from plinder.core.split.utils import get_split
 from plinder.core.utils import cpl
 from plinder.core.utils.config import get_config
 from plinder.core.utils.log import setup_logger
@@ -51,7 +50,8 @@ def query_index(
     df = sql(query).to_df()
     if splits is None:
         splits = ["train", "val"]
-    split_df = get_split(cfg=cfg)
+    split = cpl.get_plinder_path(rel=f"{cfg.data.splits}/{cfg.data.split_file}")
+    split_df = pd.read_parquet(split)
     split_dict = dict(zip(split_df["system_id"], split_df["split"]))
     df["split"] = df["system_id"].map(lambda x: split_dict.get(x, "unassigned"))
     if "*" not in splits:

src/plinder/core/split/plot.py

@@ -166,7 +166,7 @@ def from_files(
             mms_count=mms_count,
         )
         if plindex_file is None:
-            plindex = get_plindex()
+            plindex = get_plindex().drop(columns=["split"])
         else:
             plindex = pd.read_parquet(plindex_file)
         plotter.plindex = plotter.merge_splits_and_plindex(plindex)

src/plinder/data/column_descriptions/ligands.tsv

@@ -7,7 +7,7 @@ ligand_bird_id	str	Ligand BIRD id
 ligand_centroid	list[float]	Ligand center of geometry
 ligand_smiles	str	Ligand SMILES based on OpenStructure dictionary lookup, or resolved SMILES if not in dictionary
 ligand_resolved_smiles	str	SMILES of only resolved ligand atoms
-ligand_rdkit_canonical_smiles	str | None	RDKit canonical SMILES
+ligand_rdkit_canonical_smiles	str | None	RDKit canonical SMILES (Recommended)
 ligand_molecular_weight	float | None	Molecular weight
 ligand_crippen_clogp	float | None	Ligand Crippen MlogP, see https://www.rdkit.org/docs/source/rdkit.Chem.Crippen.html
 ligand_num_rot_bonds	int | None	Number of rotatable bonds