化合物をダウンロードして分類・回帰する：chembl-webresource-client × RDKit × RandomForest #Python

from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
from sklearn.metrics import classification_report, accuracy_score, mean_squared_error, r2_score

def regress_molecules(df_dict, key):
    """
    指定された結合タンパク質の化合物のSMILESをECFP4に変換し、
    standard_value, value, log(standard_value), log(value) に対する回帰モデルを構築し、評価を行う。
    
    Parameters:
        df_dict (dict): key が結合タンパク質の名前、value が SMILES・目的変数を含む DataFrame
        key (str): 対象の結合タンパク質の名前
    """
    if key not in df_dict:
        raise ValueError(f"Key '{key}' not found in df_dict.")
    
    df = df_dict[key].dropna(subset=["canonical_smiles", "standard_value", "value"])
    
    fp_gen = GetMorganGenerator(radius=2, fpSize=1024)
    
    data, target_std, target_val = [], [], []
    for smi, std, val in zip(df["canonical_smiles"], df["standard_value"], df["value"]):
        mol = Chem.MolFromSmiles(smi)
        if mol:
            fp = np.array(fp_gen.GetFingerprint(mol))
            data.append(fp)
            target_std.append(float(std))
            target_val.append(float(val))
    
    # 目的変数リスト（NaNや無効値を除外）
    targets = {
        "standard_value": np.array(target_std),
        "value": np.array(target_val),
        "log_standard_value": np.log1p(target_std),
        "log_value": np.log1p(target_val)
    }
    
    # 特徴量をNumPy配列に変換
    data = np.array(data)
    
    # 各目的変数に対して回帰モデルを構築
    results = {}
    for name, target in targets.items():
        valid_idx = np.isfinite(target)  # 有効なデータのみを使用
        X_valid = data[valid_idx]
        y_valid = target[valid_idx]
        
        X_train, X_test, y_train, y_test = train_test_split(X_valid, y_valid, test_size=0.2, random_state=42)
        
        model = RandomForestRegressor(n_estimators=100, random_state=42)
        model.fit(X_train, y_train)
        
        y_pred = model.predict(X_test)
        mse = mean_squared_error(y_test, y_pred)
        r2 = r2_score(y_test, y_pred)
        
        results[name] = {"MSE": mse, "R2": r2}
        
    # 結果表示
    print("R2 Scores: {:.3f} (standard_value)\t{:.3f} (value) \t{:.3f} (log_standard_value) \t{:.3f} (log_value)\t{}".format(
            results['standard_value']['R2'],
            results['value']['R2'],
            results['log_standard_value']['R2'],
            results['log_value']['R2'],
            key
        ))