Defective SMILES strings in pcqm4mv2

[0xnurl]

When loading pcqm4mv2 using the provided dataloader, the following error occurs because of defective SMILES strings: (see #498)

'NoneType' object has no attribute 'GetAtoms'

Using RDKit 2025.9.2.

It seems that newer versions of RDKit enforces things more strictly.

These are the defective strings and their index in the CSV:

51128, O[Si]123O[Si]3(O1)(O2)O
142541, C1[SiH2]2[SiH2]1C2
155174, Cl[SiH]12C[SiH2]2C1
155175, Cl[SiH]12C[SiH]2(C1)Cl
155176, Cl[Si]12(Cl)C[SiH]2(C1)Cl
155180, c1ccc(cc1)[SiH]12C[SiH]2(C1)c1ccccc1
155181, Br[SiH]12C[SiH2]2C1
193600, C[Si]1234C[Si]564C42[Si]23(C1)[Si]54(C2)(C6)C
411357, CO[Si]12(C)O[Si]2(O1)(C)OC
505556, CC1(C)[Si]2[Si]1(C2)(C)C
1756391, CCO[Si]12(C)C[Si]2(C1)(C)OCC
1756392, CC1[Si]2([Si]1(C2)(Cl)Cl)(Cl)Cl
2064546, c1ccc(cc1)[Si]12(C[Si]2C1)c1ccccc1
2100882, CO[Si]1(OC)(OC)N(C)[C](N1C)N(C)C
2119947, C[Si]123O[Si]3(O1)(O2)C
2539039, CN1[C](O[P]1(F)(F)(F)F)c1ccccc1
2620153, [O]N1O[Si]2(O1)([O])ON(O2)[O]
2661113, F[Si](F)(F)(F)(F)F
2690488, Cl[C]1C=CC=C2N1C[Si](O2)(C)(C)Cl
2690489, C[C]1C=CC=C2N1C[Si](O2)(C)(C)Cl
2885109, CO[Si]12(C)C[Si]2(C1)(OC)OC
2885110, CO[Si]12(OC)C[Si]2(C1)(OC)OC
2980839, F[P](F)(F)(F)(F)F
2992920, F[Si]12(F)C[Si]2(C1)(F)F
3003839, C[Si]1(C)(Cl)OC2=CC=C[CH]N2C1
3006497, C[Si]12(C[Si]2(C1)(C)C)c1ccccc1
3096168, C[Si]12(C)C[Si]2C1
3132431, C[Si]12(C)C[Si]2(C1)(Cl)Cl
3132432, C[Si]12(Cl)C[Si]2(C1)(C)Cl
3143920, Fc1ccc(cc1)[C]1OC[Si](O1)(F)(F)F
3143921, Cc1ccc(cc1)[C]1OC[Si](O1)(F)(F)F
3208535, C=C[Si]12(C)C[Si]2(C1)(C)C
3318575, ClC1=CC=C2N([CH]1)C[Si](O2)(C)(C)Cl
3407347, C[Si]12(Cl)C[Si]2(C1)(C)C
3432361, O[Si]1(O)O[Si]23([Si](O1)(O2)(O3)O)O
3562751, Cl[Si]12(Cl)C[Si]2(C1)(Cl)Cl
3648958, C=C[Si]12(C)C[Si]2(C1)(C)C=C
3694349, CO[Si]12(OC)O[Si]2(O1)(OC)OC
3714998, [O]P123OP45([Si]3(O1)(O2)(O4)O5)[O]

To (temporarily) fix the issue, you can run the following script:

import pandas as pd
from rdkit import Chem
from tqdm import tqdm

filename = "data.csv"
df = pd.read_csv(filename)

idxs_to_remove = []

for index, row in tqdm(df.iterrows()):
    mol_generated = Chem.MolFromSmiles(row["smiles"])
    if mol_generated is None:
        idxs_to_remove.append(row["idx"])
        print(f'{row["idx"]}, {row["smiles"]}')


df_cleaned = df[~df["idx"].isin(idxs_to_remove)]

df_cleaned.to_csv("data_clean.csv", index=False)

Then run:

cp data_clean.csv data.csv # Overwrites the original file
gzip data.csv 

And copy data.csv.gz to dataset/pcqm4m-v2/raw.