Added random split for non-SMILES identifiers

data/tabular/formation_energies/meta.yaml

@@ -22,7 +22,7 @@ targets:
 benchmarks: []
 identifiers:
     - id: composition
-      type: text
+      type: COMPOSITION
       description: chemical formula
 license: CC BY 4.0
 links:

data/tabular/h2_storage_materials/meta.yaml

@@ -19,7 +19,7 @@ identifiers:
       type: IUPAC
       description: chemical name
     - id: chemical_formula
-      type: Other
+      type: COMPOSITION
       names:
           - noun: chemical formula
       description: chemical formula

data/tabular/mp_anisotropy/meta.yaml

@@ -19,7 +19,7 @@ benchmarks:
       split_column: split
 identifiers:
     - id: formula
-      type: text
+      type: COMPOSITION
       description: composition
 license: CC BY 4.0
 links:

data/tabular/mp_shear_modulus/meta.yaml

@@ -19,7 +19,7 @@ benchmarks:
       split_column: split
 identifiers:
     - id: formula
-      type: text
+      type: COMPOSITION
       description: composition
 license: CC BY 4.0
 links:

data/tabular/peptides_hemolytic/meta.yaml

@@ -24,7 +24,7 @@ targets:
 benchmarks: []
 identifiers:
     - id: sequence
-      type: Other
+      type: AS_SEQUENCE
       description: amino acid sequence
 license: CC BY 4.0
 links:

data/tabular/peptides_nonfouling/meta.yaml

@@ -2,13 +2,13 @@
 name: peptides_nonfouling
 description: |-
     Non-fouling is defined as resistance to non-specific interactions.
-            A non-fouling peptide (positive example) is defined using the mechanism proposed in
-            ref white2012decoding. Briefly, ref white2012decoding, showed that the exterior surfaces
-            of proteins have a significantly different frequency of amino acids, and this increases
-            in aggregation prone environments, like the cytoplasm. Synthesizing self-assembling peptides
-            that follow this amino acid distribution and coating surfaces with the peptides creates
-            non-fouling surfaces. This pattern was also found inside chaperone proteins,
-            another area where resistance to non-specific interactions is important (ref white2012role).
+    A non-fouling peptide (positive example) is defined using the mechanism proposed in
+    ref white2012decoding. Briefly, ref white2012decoding, showed that the exterior surfaces
+    of proteins have a significantly different frequency of amino acids, and this increases
+    in aggregation prone environments, like the cytoplasm. Synthesizing self-assembling peptides
+    that follow this amino acid distribution and coating surfaces with the peptides creates
+    non-fouling surfaces. This pattern was also found inside chaperone proteins,
+    another area where resistance to non-specific interactions is important (ref white2012role).
 targets:
     - id: nonfouling
       description: The nonfouling activity of a peptide sequence (1) or not (0).
@@ -21,7 +21,7 @@ targets:
 benchmarks: []
 identifiers:
     - id: sequence
-      type: Other
+      type: AS_SEQUENCE
       description: amino acid sequence
 license: CC BY 4.0
 links:

data/tabular/peptides_soluble/meta.yaml

@@ -2,12 +2,12 @@
 name: peptides_soluble
 description: |-
     Solubility was estimated by retrospective analysis of electronic laboratory notebooks.
-            The notebooks were part of a large effort called the Protein Structure Initiative and consider sequences
-            linearly through the following stages: Selected,  Cloned,  Expressed,  Soluble,  Purified, Crystallized,
-            HSQC (heteronuclear single quantum coherence), Structure, and deposited in PDB. The peptides were identified
-            as soluble or insoluble by "Comparing the experimental status at two time points, September 2009 and May 2010,
-            we were able to derive a set of insoluble proteins defined as those which were not
-            soluble in September 2009 and still remained in that state 8 months later."
+    The notebooks were part of a large effort called the Protein Structure Initiative and consider sequences
+    linearly through the following stages: Selected,  Cloned,  Expressed,  Soluble,  Purified, Crystallized,
+    HSQC (heteronuclear single quantum coherence), Structure, and deposited in PDB. The peptides were identified
+    as soluble or insoluble by "Comparing the experimental status at two time points, September 2009 and May 2010,
+    we were able to derive a set of insoluble proteins defined as those which were not
+    soluble in September 2009 and still remained in that state 8 months later."
 targets:
     - id: soluble
       description: The solubility of a peptide sequence (1) or not (0).
@@ -20,7 +20,7 @@ targets:
 benchmarks: []
 identifiers:
     - id: sequence
-      type: Other
+      type: AS_SEQUENCE
       description: amino acid sequence
 license: CC BY 4.0
 links:

data/tabular/perovskite_db/meta.yaml

@@ -46,7 +46,7 @@ targets:
 benchmarks: []
 identifiers:
     - id: reduced_formulas
-      type: Other
+      type: COMPOSITION
       description: reduced chemical formula
     - id: iupac_formulas
       type: Other

data/tabular/rdkit_features/meta.yaml

@@ -1,5 +1,5 @@
 ---
-name: rdkit_features
+names: rdkit_features
 description: |-
     Molecular descriptors computed using RDKit
 targets:

data/tabular/scaffold_split.py

@@ -0,0 +1,182 @@
+from glob import glob
+import yaml
+import pandas as pd
+from chemnlp.data.split import _create_scaffold_split
+import fire
+import os
+import subprocess
+from pandarallel import pandarallel
+from typing import List
+from functools import partial
+
+pandarallel.initialize(progress_bar=True)
+
+to_scaffold_split = [
+    "blood_brain_barrier_martins_et_al",
+    "serine_threonine_kinase_33_butkiewicz",
+    "ld50_zhu",
+    "herg_blockers",
+    "clearance_astrazeneca",
+    "half_life_obach",
+    "drug_induced_liver_injury",
+    "kcnq2_potassium_channel_butkiewicz",
+    "sarscov2_3clpro_diamond",
+    "volume_of_distribution_at_steady_state_lombardo_et_al",
+    "sr_hse_tox21",
+    "nr_er_tox21",
+    "cyp2c9_substrate_carbonmangels",
+    "nr_aromatase_tox21",
+    "cyp_p450_2d6_inhibition_veith_et_al",
+    "cyp_p450_1a2_inhibition_veith_et_al",
+    "cyp_p450_2c9_inhibition_veith_et_al",
+    "m1_muscarinic_receptor_antagonists_butkiewicz",
+    "nr_ar_tox21",
+    "sr_atad5_tox21",
+    "tyrosyl-dna_phosphodiesterase_butkiewicz",
+    "cav3_t-type_calcium_channels_butkiewicz",
+    "clintox" "sr_p53_tox21",
+    "nr_er_lbd_tox21" "pampa_ncats",
+    "sr_mmp_tox21",
+    "caco2_wang",
+    "sarscov2_vitro_touret",
+    "choline_transporter_butkiewicz",
+    "orexin1_receptor_butkiewicz",
+    "human_intestinal_absorption",
+    "nr_ahr_tox21",
+    "cyp3a4_substrate_carbonmangels",
+    "herg_karim_et_al",
+    "hiv",
+    "carcinogens",
+    "sr_are_tox21",
+    "nr_ppar_gamma_tox21",
+    "solubility_aqsoldb",
+    "m1_muscarinic_receptor_agonists_butkiewicz",
+    "ames_mutagenicity",
+    "potassium_ion_channel_kir2_1_butkiewicz",
+    "cyp_p450_3a4_inhibition_veith_et_al",
+    "skin_reaction",
+    "cyp2d6_substrate_carbonmangels",
+    "cyp_p450_2c19_inhibition_veith_et_al",
+    "nr_ar_lbd_tox21",
+    "p_glycoprotein_inhibition_broccatelli_et_al",
+    "bioavailability_ma_et_al",
+    "BACE",
+    "hiv",
+    "lipophilicity",
+    "thermosol",
+    "MUV_466",
+    "MUV_548",
+    "MUV_600",
+    "MUV_644",
+    "MUV_652",
+    "MUV_689",
+    "MUV_692",
+    "MUV_712",
+    "MUV_713",
+    "MUV_733",
+    "MUV_737",
+    "MUV_810",
+    "MUV_832",
+    "MUV_846",
+    "MUV_852",
+    "MUV_858",
+    "MUV_859",
+]
+
+
+def split_for_smiles(smiles: str, train_smiles: List[str], val_smiles: List[str]):
+    if smiles in train_smiles:
+        return "train"
+    elif smiles in val_smiles:
+        return "valid"
+    else:
+        return "test"
+
+
+def main(
+    data_dir,
+    override: bool = False,
+    train_frac: float = 0.7,
+    val_frac: float = 0.15,
+    test_frac: float = 0.15,
+    seed: int = 42,
+    debug: bool = False,
+):
+    all_yaml_files = glob(os.path.join(data_dir, "**", "*.yaml"), recursive=True)
+    if debug:
+        all_yaml_files = all_yaml_files[:5]
+    transformed_files = []
+    for file in all_yaml_files:
+        with open(file, "r") as f:
+            meta = yaml.safe_load(f)
+
+        if meta["name"] in to_scaffold_split:
+            # run transform.py script in this directory if `data_clean.csv` does not exist
+            # use this dir as cwd
+            if not os.path.exists(
+                os.path.join(os.path.dirname(file), "data_clean.csv")
+            ):
+                subprocess.run(
+                    ["python", "transform.py"],
+                    cwd=os.path.dirname(file),
+                )
+
+            transformed_files.append(
+                os.path.join(os.path.dirname(file), "data_clean.csv")
+            )
+
+    all_smiles = set()
+    for file in transformed_files:
+        df = pd.read_csv(file)
+        all_smiles.update(df["SMILES"].tolist())
+        del df  # ensure memory is freed
+
+    all_smiles = list(all_smiles)
+    splits = _create_scaffold_split(
+        all_smiles, frac=[train_frac, val_frac, test_frac], seed=42
+    )
+
+    # select the right indices for each split
+    train_smiles = [all_smiles[i] for i in splits["train"]]
+    val_smiles = [all_smiles[i] for i in splits["valid"]]
+    print(
+        "Train smiles:",
+        len(train_smiles),
+        "Val smiles:",
+        len(val_smiles),
+        "Test smiles:",
+        len(splits["test"]),
+    )
+
+    # now for each dataframe, add a split column based on in which list in the `splits` dict the smiles is
+    # smiles are in the SMILES column
+    # if the override option is true, we write this new file to `data_clean.csv` in the same directory
+    # otherwise, we copy the old `data_clean.csv` to `data_clean_old.csv` and write the new file to `data_clean.csv`
+    # we print the train/val/test split sizes and fractions for each dataset
+
+    split_for_smiles_curried = partial(
+        split_for_smiles, train_smiles=train_smiles, val_smiles=val_smiles
+    )
+    for file in transformed_files:
+        df = pd.read_csv(file)
+        df["split"] = df["SMILES"].parallel_apply(split_for_smiles_curried)
+
+        # to ensure overall scaffold splitting does not distort train/val/test split sizes for each dataset
+        print(
+            f"Dataset {file} has {len(df)} datapoints. Split sizes: {len(df[df['split'] == 'train'])} train, {len(df[df['split'] == 'valid'])} valid, {len(df[df['split'] == 'test'])} test."  # noqa: E501
+        )
+        print(
+            f"Dataset {file} has {len(df)} datapoints. Split fractions: {len(df[df['split'] == 'train']) / len(df)} train, {len(df[df['split'] == 'valid']) / len(df)} valid, {len(df[df['split'] == 'test']) / len(df)} test."  # noqa: E501
+        )
+        if override:
+            # write the new data_clean.csv file
+            df.to_csv(file, index=False)
+        else:
+            # copy the old data_clean.csv to data_clean_old.csv
+            os.rename(file, file.replace(".csv", "_old.csv"))
+            # write the new data_clean.csv file
+            df.to_csv(file, index=False)
+
+
+if __name__ == "__main__":
+    fire.Fire(main)