Upload dataset and scripts

README.md

@@ -1,3 +1,54 @@
----
-license: mit
----
+---
+license: mit
+pretty_name: Drug-Likeness RDKit Dataset
+tags:
+  - cheminformatics
+  - rdkit
+  - drug-likeness
+  - classification
+  - AutoML
+task_categories:
+  - classification
+task_ids:
+  - binary-classification
+source_datasets:
+  - DBPP-Predictor(https://github.com/yxgu2353/DBPP-Predictor.git)
+---
+
+# Drug-Likeness Prediction Dataset (Based on DBPP-Predictor Data)
+
+This dataset was created as part of a final project on drug-likeness prediction, based on the data from the DBPP-Predictor paper:
+
+> Gu, Y., Wang, Y., Zhu, K. et al. DBPP-Predictor: a novel strategy for prediction of chemical drug-likeness based on property profiles. J Cheminform 16, 4 (2024). https://doi.org/10.1186/s13321-024-00800-9
+
+ It includes curated molecular data, preprocessed RDKit descriptors, and training/test splits suitable for training classification models to distinguish drug-like from non-drug-like molecules.
+
+## Project Background
+
+Drug-likeness refers to the potential of a small molecule to become a drug. Traditional rule-based approaches (e.g., Lipinski's Rule of Five) often fail to generalize across complex compounds. This project uses RDKit descriptors and AutoML (H2O) to construct a highly interpretable, generalizable classification model.
+
+## Dataset Description
+
+The dataset is derived from the DBPP GitHub repository (https://github.com/yxgu2353/DBPP-Predictor.git). It includes:
+- 5,147 drug-like molecules (FDA and globally approved drugs)
+- 10,000 non-drug-like molecules sampled from ZINC
+
+All molecules were standardized using MolVS(clean_data.py), the datasets were split by sklearn(split_dataset.py), and RDKit descriptors (216 features) were computed (Rdkit_descriptor.py).
+
+### Files:
+- 'train.parquet': Raw training set (SMILES + labels)
+- 'test.parquet': Raw test set (SMILES + labels)
+- 'train_rdkit_descriptors.parquet': RDKit descriptors for training data
+- 'test_rdkit_descriptors.parquet': RDKit descriptors for test data
+
+Each descriptor file contains:
+- 'label': 0 for non-drug, 1 for drug
+- 'Standardized_SMILES'
+- 216 numerical RDKit descriptors
+
+## Model Development
+
+All models were trained using H2O AutoML with 10-fold cross-validation(model_constrcution.py). The top 3 models were also evaluated using an independent test set, and SHAP analysis was used to interpret the top structural features contributing to drug-likeness(model_analysis.py).
+
+
+

scripts/.virtual_documents/model_construction.ipynb

@@ -0,0 +1,187 @@
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import h2o
+from h2o.automl import H2OAutoML
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+
+h2o.init()
+
+
+df_train = pd.read_csv("./intermediate/train_rdkit_descriptors.csv")
+df_test = pd.read_csv("./intermediate/test_rdkit_descriptors.csv")
+
+
+x_train = df_train.drop(columns=["label", "Standardized_SMILES"])
+y_train = df_train["label"]
+
+
+x_test = df_test.drop(columns=["label", "Standardized_SMILES"])
+y_test = df_test["label"]
+
+
+train_h2o = h2o.H2OFrame(pd.concat([x_train, y_train], axis=1))
+test_h2o = h2o.H2OFrame(pd.concat([x_test, y_test], axis=1))
+
+
+train_h2o["label"] = train_h2o["label"].asfactor()
+test_h2o["label"] = test_h2o["label"].asfactor()
+
+
+feature_cols = x_train.columns.tolist()
+
+
+aml = H2OAutoML(max_models=20,seed=42,nfolds=10,sort_metric="AUC")
+aml.train(x=feature_cols, y="label", training_frame=train_h2o)
+
+
+lb = aml.leaderboard
+lb.head(rows=lb.nrows)
+best_model = aml.leader
+
+
+top_3_models = lb.as_data_frame()["model_id"].head(3).tolist()
+
+
+top_3_models
+
+
+for i, model_id in enumerate(top_3_models, start=1):
+    model = h2o.get_model(model_id)
+    model_path = h2o.save_model(model=model, path=f"../product/top_model_{i}", force=True)
+
+
+model_ids = lb.as_data_frame()["model_id"].tolist()
+all_model_summaries = []
+for model_id in model_ids:
+    model = h2o.get_model(model_id)
+    cv_summary = model.cross_validation_metrics_summary().as_data_frame()
+    cv_summary["model_id"] = model_id
+    all_model_summaries.append(cv_summary)
+
+cv_all_models_df = pd.concat(all_model_summaries, ignore_index=True)
+cv_all_models_df.to_csv("../intermediate/cross_validation_result.csv",index=False)
+
+
+
+
+
+from h2o import load_model
+restored_model1 = h2o.load_model("./product/top_model_1/StackedEnsemble_AllModels_1_AutoML_1_20250401_220205")
+restored_model2 = h2o.load_model("./product/top_model_2/StackedEnsemble_BestOfFamily_1_AutoML_1_20250401_220205")
+restored_model3 = h2o.load_model("./product/top_model_3/GBM_4_AutoML_1_20250401_220205")
+
+
+perf1=restored_model1.model_performance(test_h2o)
+perf2=restored_model2.model_performance(test_h2o)
+perf3=restored_model3.model_performance(test_h2o)
+
+
+
+
+
+threshold = 0.5
+acc = perf1.accuracy(thresholds=[threshold])[0][1]
+f1 = perf1.F1(thresholds=[threshold])[0][1]
+prec = perf1.precision(thresholds=[threshold])[0][1]
+rec = perf1.recall(thresholds=[threshold])[0][1]
+spec = perf1.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+
+metrics1 = {
+    "AUC": perf1.auc(),
+    "LogLoss": perf1.logloss(),
+    "Accuracy": perf1.accuracy(),
+    "F1": perf1.F1(),
+    "Precision": perf1.precision(),
+    "Recall": perf1.recall(),
+    "Specificity": perf1.specificity()
+}
+metrics1
+
+
+perf1.plot(type="roc")
+perf1.plot(type="pr")
+
+
+
+
+
+threshold = 0.5
+acc = perf2.accuracy(thresholds=[threshold])[0][1]
+f1 = perf2.F1(thresholds=[threshold])[0][1]
+prec = perf2.precision(thresholds=[threshold])[0][1]
+rec = perf2.recall(thresholds=[threshold])[0][1]
+spec = perf2.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+
+metrics2 = {
+    "AUC": perf2.auc(),
+    "LogLoss": perf2.logloss(),
+    "Accuracy": perf2.accuracy(),
+    "F1": perf2.F1(),
+    "Precision": perf2.precision(),
+    "Recall": perf2.recall(),
+    "Specificity": perf2.specificity()
+}
+
+
+metrics2
+
+
+perf2.plot(type="roc")
+perf2.plot(type="pr")
+
+
+
+
+
+threshold = 0.5
+acc = perf3.accuracy(thresholds=[threshold])[0][1]
+f1 = perf3.F1(thresholds=[threshold])[0][1]
+prec = perf3.precision(thresholds=[threshold])[0][1]
+rec = perf3.recall(thresholds=[threshold])[0][1]
+spec = perf3.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+
+metrics3 = {
+    "AUC": perf3.auc(),
+    "LogLoss": perf3.logloss(),
+    "Accuracy": perf3.accuracy(),
+    "F1": perf3.F1(),
+    "Precision": perf3.precision(),
+    "Recall": perf3.recall(),
+    "Specificity": perf3.specificity()
+}
+
+
+metrics3
+
+
+perf3.plot(type="roc")
+perf3.plot(type="pr")
+
+
+restored_model3.shap_summary_plot(test_h2o[:,:-1])
+fig = plt.gcf()
+fig.savefig("./product/3shap_summary_plot.png", dpi=300, bbox_inches="tight")

scripts/Rdkit_descriptor.py

@@ -0,0 +1,27 @@
+import pandas as pd
+import numpy as np
+from rdkit import Chem
+from rdkit.Chem import Descriptors
+
+df_train = pd.read_csv("./intermediate/train.csv")
+df_test = pd.read_csv("./intermediate/test.csv")
+df_train = df_train.drop(['SMILES'], axis=1)
+df_test = df_test.drop(['SMILES'], axis=1)
+
+descriptor_names = [desc_name for desc_name, _ in Descriptors.descList]
+len(descriptor_names)
+
+def calculate_rdkit_descriptors(smiles):
+    mol = Chem.MolFromSmiles(smiles)
+    if mol is None:
+        return None  
+    return {desc_name: func(mol) for desc_name, func in Descriptors.descList}
+
+df_train_descriptors = df_train["Standardized_SMILES"].apply(calculate_rdkit_descriptors)
+df_test_descriptors = df_test["Standardized_SMILES"].apply(calculate_rdkit_descriptors)
+df_train_descriptors = pd.DataFrame(df_train_descriptors.tolist())
+df_test_descriptors = pd.DataFrame(df_test_descriptors.tolist())
+df_train_final = pd.concat([df_train, df_train_descriptors], axis=1)
+df_test_final = pd.concat([df_test, df_test_descriptors], axis=1)
+df_train_final.to_csv("./intermediate/train_rdkit_descriptors.csv", index=False)
+df_test_final.to_csv("./intermediate/test_rdkit_descriptors.csv", index=False)

scripts/clean_data.py

@@ -0,0 +1,24 @@
+from rdkit import Chem
+from molvs import Standardizer
+import pandas as pd
+df0 = pd.read_csv("data/negative.csv")
+df1 = pd.read_csv("data/positive.csv")
+s = Standardizer()
+
+def standardize_smiles(smiles):
+    try:
+        mol = Chem.MolFromSmiles(smiles)  
+        if mol:
+            clean_mol = s.fragment_parent(mol)  
+            standardized_mol = s.standardize(clean_mol)  
+            return Chem.MolToSmiles(standardized_mol)  
+    except:
+        return None
+
+df0["Standardized_SMILES"] = df0.iloc[:,0].apply(standardize_smiles)
+df0 = df0.dropna(subset=["Standardized_SMILES"])
+df0.to_csv("intermediate/cleaned_negative.csv", index=False)
+
+df1["Standardized_SMILES"] = df1.iloc[:,0].apply(standardize_smiles)
+df1 = df1.dropna(subset=["Standardized_SMILES"])
+df1.to_csv("intermediate/cleaned_positive.csv", index=False)

scripts/model_analysis.py

@@ -0,0 +1,121 @@
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import h2o
+from h2o.automl import H2OAutoML
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+h2o.init()
+
+###Data preparation
+df_train = pd.read_csv("./intermediate/train_rdkit_descriptors.csv")
+df_test = pd.read_csv("./intermediate/test_rdkit_descriptors.csv")
+
+x_train = df_train.drop(columns=["label",  "Standardized_SMILES"])
+y_train = df_train["label"]
+x_test = df_test.drop(columns=["label", "Standardized_SMILES"])
+y_test = df_test["label"]
+
+train_h2o = h2o.H2OFrame(pd.concat([x_train, y_train], axis=1))
+test_h2o = h2o.H2OFrame(pd.concat([x_test, y_test], axis=1))
+
+train_h2o["label"] = train_h2o["label"].asfactor()
+test_h2o["label"] = test_h2o["label"].asfactor()
+feature_cols = x_train.columns.tolist()
+
+###Reload the model
+from h2o import load_model
+restored_model1 = h2o.load_model("../product/top_model_1/StackedEnsemble_AllModels_1_AutoML_1_20250401_220205")
+restored_model2 = h2o.load_model("../product/top_model_2/StackedEnsemble_BestOfFamily_1_AutoML_1_20250401_220205")
+restored_model3 = h2o.load_model("../product/top_model_3/GBM_4_AutoML_1_20250401_220205")
+
+###Test the model
+perf1=restored_model1.model_performance(test_h2o)
+perf2=restored_model2.model_performance(test_h2o)
+perf3=restored_model3.model_performance(test_h2o)
+
+###Get the result with different threshold
+
+threshold = 0.5
+acc = perf1.accuracy(thresholds=[threshold])[0][1]
+f1 = perf1.F1(thresholds=[threshold])[0][1]
+prec = perf1.precision(thresholds=[threshold])[0][1]
+rec = perf1.recall(thresholds=[threshold])[0][1]
+spec = perf1.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+threshold = 0.5
+acc = perf2.accuracy(thresholds=[threshold])[0][1]
+f1 = perf2.F1(thresholds=[threshold])[0][1]
+prec = perf2.precision(thresholds=[threshold])[0][1]
+rec = perf2.recall(thresholds=[threshold])[0][1]
+spec = perf2.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+threshold = 0.5
+acc = perf3.accuracy(thresholds=[threshold])[0][1]
+f1 = perf3.F1(thresholds=[threshold])[0][1]
+prec = perf3.precision(thresholds=[threshold])[0][1]
+rec = perf3.recall(thresholds=[threshold])[0][1]
+spec = perf3.specificity(thresholds=[threshold])[0][1]
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+metrics1 = {
+    "AUC": perf1.auc(),
+    "LogLoss": perf1.logloss(),
+    "Accuracy": perf1.accuracy(),        
+    "F1": perf1.F1(),
+    "Precision": perf1.precision(),
+    "Recall": perf1.recall(),
+    "Specificity": perf1.specificity()    
+}
+
+metrics2 = {
+    "AUC": perf2.auc(),
+    "LogLoss": perf2.logloss(),
+    "Accuracy": perf2.accuracy(),
+    "F1": perf2.F1(),
+    "Precision": perf2.precision(),
+    "Recall": perf2.recall(),
+    "Specificity": perf2.specificity()   
+}
+
+metrics3 = {
+    "AUC": perf3.auc(),
+    "LogLoss": perf3.logloss(),
+    "Accuracy": perf3.accuracy(),
+    "F1": perf3.F1(),
+    "Precision": perf3.precision(),
+    "Recall": perf3.recall(),
+    "Specificity": perf3.specificity()   
+}
+###Get the figure for roc and pr
+perf1.plot(type="roc")
+perf1.plot(type="pr")
+
+perf2.plot(type="roc")
+perf2.plot(type="pr")
+
+perf3.plot(type="roc")
+perf3.plot(type="pr")
+
+### SHAP analysis
+restored_model3.shap_summary_plot(test_h2o[:,:-1])
+fig = plt.gcf()
+fig.savefig("./product/3shap_summary_plot.png", dpi=300, bbox_inches="tight")

scripts/model_constrcution.py

@@ -0,0 +1,87 @@
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+import h2o
+from h2o.automl import H2OAutoML
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+h2o.init()
+
+###Data preparation
+df_train = pd.read_csv("./intermediate/train_rdkit_descriptors.csv")
+df_test = pd.read_csv("./intermediate/test_rdkit_descriptors.csv")
+
+x_train = df_train.drop(columns=["label",  "Standardized_SMILES"])
+y_train = df_train["label"]
+x_test = df_test.drop(columns=["label", "Standardized_SMILES"])
+y_test = df_test["label"]
+
+train_h2o = h2o.H2OFrame(pd.concat([x_train, y_train], axis=1))
+test_h2o = h2o.H2OFrame(pd.concat([x_test, y_test], axis=1))
+
+train_h2o["label"] = train_h2o["label"].asfactor()
+test_h2o["label"] = test_h2o["label"].asfactor()
+feature_cols = x_train.columns.tolist()
+
+###Training
+aml = H2OAutoML(max_models=20,seed=42,nfolds=10,sort_metric="AUC")
+aml.train(x=feature_cols, y="label", training_frame=train_h2o)
+
+###Model Selection
+lb = aml.leaderboard
+lb.head(rows=lb.nrows)
+best_model = aml.leader
+
+###Save the best model
+#best_model = h2o.save_model(model = best_model, path ='./product/', force = True)
+top_3_models = lb.as_data_frame()["model_id"].head(3).tolist()
+
+for i, model_id in enumerate(top_3_models, start=1):
+    model = h2o.get_model(model_id)
+    model_path = h2o.save_model(model=model, path=f"./product/top_model_{i}", force=True)
+
+###Save the corss validation result for all models
+model_ids = lb.as_data_frame()["model_id"].tolist()
+all_model_summaries = []
+for model_id in model_ids:
+    model = h2o.get_model(model_id)
+    cv_summary = model.cross_validation_metrics_summary().as_data_frame()
+    cv_summary["model_id"] = model_id
+    all_model_summaries.append(cv_summary)
+
+cv_all_models_df = pd.concat(all_model_summaries, ignore_index=True)
+cv_all_models_df.to_csv("./intermediate/cross_validation_result.csv",index=False)
+
+###Reload the model for addtional test
+from h2o import load_model
+restored_model1 = h2o.load_model("./product/top_model_1/StackedEnsemble_AllModels_1_AutoML_1_20250401_220205")
+restored_model2 = h2o.load_model("./product/top_model_2/StackedEnsemble_BestOfFamily_1_AutoML_1_20250401_220205")
+restored_model3 = h2o.load_model("./product/top_model_3/GBM_4_AutoML_1_20250401_220205")
+
+perf1=restored_model1.model_performance(test_h2o)
+perf2=restored_model2.model_performance(test_h2o)
+perf3=restored_model3.model_performance(test_h2o)
+
+###Obtain the performance
+threshold = 0.5
+acc = perf1.accuracy(thresholds=[threshold])[0][1]
+f1 = perf1.F1(thresholds=[threshold])[0][1]
+prec = perf1.precision(thresholds=[threshold])[0][1]
+rec = perf1.recall(thresholds=[threshold])[0][1]
+spec = perf1.specificity(thresholds=[threshold])[0][1]
+AUC= perf1.auc()
+LogLoss=perf1.logloss()
+print(f"AUC = {AUC}")
+print(f"LogLoss = {LogLoss}")
+print(f"Threshold = {threshold}")
+print(f"Accuracy     = {acc:.4f}")
+print(f"F1 Score     = {f1:.4f}")
+print(f"Precision    = {prec:.4f}")
+print(f"Recall       = {rec:.4f}")
+print(f"Specificity  = {spec:.4f}")
+
+###SHAP analysis for the third-best model
+restored_model3.shap_summary_plot(test_h2o[:,:-1])
+fig = plt.gcf()
+fig.savefig("./product/3shap_summary_plot.png", dpi=300, bbox_inches="tight")

scripts/split_dataset.py

@@ -0,0 +1,16 @@
+import numpy as np
+import pandas as pd
+from sklearn.model_selection import train_test_split
+
+df_positive = pd.read_csv("./intermediate/cleaned_positive.csv")  
+df_negative = pd.read_csv("./intermediate/cleaned_negative.csv")
+
+pos_train, pos_test = train_test_split(df_positive, test_size=0.2, random_state=42)
+neg_train_full, neg_test = train_test_split(df_negative, test_size=0.2, random_state=42)
+neg_train = neg_train_full.sample(n=len(pos_train), random_state=42) 
+neg_remaining = neg_train_full.drop(neg_train.index)
+neg_test = pd.concat([neg_test, neg_remaining]).reset_index(drop=True)
+train_df = pd.concat([pos_train, neg_train]).sample(frac=1, random_state=42).reset_index(drop=True) 
+test_df = pd.concat([pos_test, neg_test]).sample(frac=1, random_state=42).reset_index(drop=True)  
+train_df.to_csv("./intermediate/train.csv", index=False)
+test_df.to_csv("./intermediate/test.csv", index=False)