Skip to content

Use case: Cheminformatics in a Python/Jupyter notebook

Problem

You want to explore a compound library in a Jupyter notebook — compute descriptors, cluster by fingerprint similarity, build a quick ML model — but setting up RDKit via conda breaks your existing environment or doesn't work in Colab.

Solution

chematic installs with pip install chematic in any Python environment: conda-free, C++-free, Colab-ready. Descriptors run in parallel Rust under the hood; mol renders a 2D structure inline without extra config.

Output / What you get

Loaded 5 000 molecules
2 312 / 5 000 pass Lipinski + PAINS
CV AUC: 0.823 ± 0.041

=== Compound_042 ===
Molecular weight 312.4 Da, formula C18H20N2O3.
LogP 2.41 (moderately lipophilic), TPSA 58.2 Ų.
HBD 1, HBA 4, 4 rotatable bonds, 2 aromatic rings.
Drug-likeness: no Lipinski rule-of-5 violations. Likely orally bioavailable.
QED 0.74. No structural alerts (PAINS / Brenk clean).
  • chematic.descriptors_df(smiles_list) — one-liner pandas DataFrame of 72 descriptors
  • chematic.bulk.ecfp4(smiles_list) — (N, 2048) uint8 for sklearn / PyTorch
  • chematic.SimilarityIndex — LSH approximate nearest-neighbour search
  • mol.svg() / chematic.depict_grid(mols, cols=3) — SVG rendering in Jupyter
  • mol.describe() — natural-language summary for reports
  • Open in Colab

Environment

pip install chematic pandas numpy scikit-learn matplotlib

1. Load a compound library from SDF

import chematic
import pandas as pd

records = list(chematic.iter_sdf("library.sdf"))
df = pd.DataFrame({
    "name":  [r.name for r in records],
    "smiles": [r.mol.smiles for r in records],
    "activity": [float(r.get("IC50_nM") or 0) for r in records],
})
print(f"Loaded {len(df)} molecules")

2. Compute 190+ descriptors in parallel

desc_df = pd.DataFrame(chematic.bulk.descriptors(df["smiles"].tolist()))
df = pd.concat([df, desc_df], axis=1)

# Quick filter
drug_like = df[df["lipinski_passes"] & df["pains_passes"]]
print(f"{len(drug_like)} / {len(df)} pass Lipinski + PAINS")

3. Generate ECFP4 fingerprints for ML

import numpy as np

X = chematic.bulk.ecfp4(df["smiles"].tolist())   # (N, 2048) uint8
y = (df["activity"] < 100).astype(int)           # binary: IC50 < 100 nM

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import cross_val_score

clf = RandomForestClassifier(n_estimators=100, random_state=42)
scores = cross_val_score(clf, X, y, cv=5)
print(f"CV AUC: {scores.mean():.3f} ± {scores.std():.3f}")

4. Tanimoto similarity clustering

import matplotlib.pyplot as plt

sim_matrix = chematic.bulk.tanimoto(
    df["smiles"].tolist(), df["smiles"].tolist()
)  # (N, N) float32

# Hierarchical clustering
from scipy.cluster.hierarchy import dendrogram, linkage
from scipy.spatial.distance import squareform

dist = 1.0 - sim_matrix
linkage_mat = linkage(squareform(dist), method="ward")
dendrogram(linkage_mat, no_labels=True)
plt.title("Compound cluster (Tanimoto distance)")
plt.show()

5. Nearest-neighbor search in a large library

# Works for hundreds of thousands of molecules
idx = chematic.SimilarityIndex.from_smiles(df["smiles"].tolist())

hits = idx.search("CC(=O)Nc1ccc(O)cc1", threshold=0.4, k=10)
for mol_idx, score in hits:
    print(f"{df['name'].iloc[mol_idx]}: Tanimoto {score:.3f}")

6. Visualize in Jupyter

from IPython.display import SVG, display

mol = chematic.from_smiles("CC(=O)Nc1ccc(O)cc1")
display(SVG(mol.svg()))

# Highlight SMARTS match
matches = chematic.smarts_find("[NH]C(=O)", mol)
atoms = [i for m in matches for i in m]
display(SVG(mol.svg_highlighted(atoms, color="#FF6B6B")))

# Grid of top hits
top_mols = [chematic.from_smiles(df["smiles"].iloc[i]) for i, _ in hits[:6]]
display(SVG(chematic.depict_grid(top_mols, cols=3)))

7. Natural language summary for reporting

for _, row in drug_like.head(3).iterrows():
    mol = chematic.from_smiles(row["smiles"])
    print(f"=== {row['name']} ===")
    print(mol.describe())
    print()

Output:

=== Compound_042 ===
Molecular weight 312.4 Da, formula C18H20N2O3.
LogP 2.41 (moderately lipophilic), TPSA 58.2 Ų.
HBD 1, HBA 4, 4 rotatable bonds, 2 aromatic rings.
Drug-likeness: no Lipinski rule-of-5 violations. Likely orally bioavailable.
QED 0.74. No structural alerts (PAINS / Brenk clean).