ALK Inhibitor Cheminformatics Pipeline

• Complete end-to-end cheminformatics and ML pipeline for analysis of ALK tyrosine kinase inhibitors from ChEMBL.
• Applied to identify structure-activity relationships and predict inhibitor potency.

Pipeline Overview

Data Preprocessing

• Merging_excel.py — dataset merging
• Remove_duplicates.py — ChEMBL ID deduplication
• Remove_lower_Ic50_duplicates.py — keeping highest IC50
• ZeroValueRemove.py — removing uninformative fingerprint bits

Fingerprint Generation

• ECFP4_caln.py — 1024-bit Morgan fingerprints
• MACC_conversion.py — MACCS key generation
• Remove95%SimilarCompounds.py — Tanimoto-based diversity filtering

Clustering

• K-meanClustering.py — silhouette-optimized K-means
• Clustering.py — Spectral clustering comparison
• kmeanAndSpectralClusteringComparison.py — cross-method analysis

Similarity & SAR Analysis

• Cluster_correlation.py — intra/inter cluster Tanimoto similarity
• cluster_corr_physicochem_euclidean_dist.py — physicochemical distances
• InClusterSimilarity.py / InClusterSimilarity_1.py — pairwise similarity
• MACC_conversion.py — fingerprint difference analysis with PCA/t-SNE
• Substructure-Based_Activity_Profiling.py — BRICS decomposition

Scaffold Analysis

• ScaffoldSearching.py — Murcko scaffold identification per cluster
• scaffold_writing_inCSV.py — scaffold export
• mcs_analysis.py — Maximum Common Substructure analysis
• MCS_all_clusterpng.py — MCS visualization
• mcs_clusters_all_in_one.py — combined MCS grid plot
• SmartstoStructure.py — SMARTS to structure conversion
• chemotype_enrichment_and_SAR_analysis.py — scaffold enrichment

Activity Prediction (ML)

• pIC50_calcn.py — IC50 to pIC50 conversion
• Diversity_and_pIC50_caln.py — similarity-filtered RF prediction
• activityPrediction_RF.py — Random Forest with PCA + GridSearchCV

Network Analysis

• NetworkGraph_commoncompounds.py — NetworkX/Plotly compound network

Visualization

• Clustering_plot.py / Clustering_plot_1.py — PCA cluster plots
• ClusterTable.py / ClusterTable_um.py — IC50 summary tables
• ColoredTable.py / ColoredTable_1.py — styled cluster tables
• IC50Range.py — IC50 distribution per cluster
• dual-axis-plot.py — KMeans vs Spectral cluster comparison
• Strct_prop_reln.py — structure-property correlations
• pharmacophore_modeling.py — HBD/HBA pharmacophore features

Methods

• Fingerprints: ECFP4 (Morgan, 1024-bit), MACCS keys
• Similarity: Tanimoto, Jaccard, cosine, Euclidean
• Clustering: K-means, Spectral (silhouette-optimized)
• Dimensionality reduction: PCA, t-SNE
• ML: Random Forest Regressor, Logistic Regression, GridSearchCV
• Scaffold: Murcko, MCS (rdFMCS), BRICS decomposition
• Network: NetworkX, Plotly interactive visualization
• Descriptors: MW, LogP, HBD, HBA, TPSA, RotBonds (RDKit)

Dependencies

pip install pandas numpy scikit-learn rdkit matplotlib seaborn plotly networkx scipy

Data Source

ChEMBL database — ALK tyrosine kinase inhibitors

Related Publications

Danishuddin, Haque M.A., Madhukar G., et al. (2025). Machine learning-driven consensus modeling for activity ranking and chemical landscape analysis of HIV-1 inhibitors. Pharmaceuticals (Basel), 18(5), 714.

Danishuddin, Haque M.A., Madhukar G., et al. (2025). Network-based clustering and statistical evaluation to elucidate structure-activity relationships of EZH2 inhibitors. SAR QSAR Environmental Research, 36(9), 827-851.

Author

Geet Madhukar, PhD Postdoctoral Research Associate University of New Hampshire geetmadhukar@gmail.com github.com/geet1210