Computational design and AI driven discovery of anaplastic lymphoma kinase inhibitors for non small cell lung cancer treatment

Abstract

Anaplastic Lymphoma Kinase (ALK), a tyrosine receptor kinase is of immense importance in non small cell lung cancer (NSCLC). Therefore, there is a need to design novel derivatives with the intention to overcome the limitations of resistance and debilitating side effects associated with current FDA-approved ALK inhibitors. This work therefore employed artificial intelligence and computational techniques to screen a library of ALK tyrosine kinase inhibitors, retrieved from ChEMBL database, with their corresponding IC50 in nM. The compounds’ descriptors were obtained using the Padel descriptors software and screened to reduce dimensionality and remove redundancy and multicollinearity. The compounds’ descriptors and their corresponding IC50 in nM were imported to Google Colab workspace with the necessary Python packages for machine learning (ML) models building. The best model was used to predict the bioactivity of new derivatives of 5-FDA approved drugs and TPX-1301, taking into account their drug-likeness properties for initials screening. The binding affinities and modes of lead compounds at the binding domain of ALK tyrosine kinase receptors were predicted using molecular docking while the binding free energies were obtained using MM-GB/SA calculations. Among all the trained models, the artificial neural network showed the most promising results, with a coefficient of determination (R2) value of 0.84, a root mean square error (RMSE) value of 0.27, a mean squared error (MSE) value of 0.22 and a mean absolute error (MAE) of 0.26. on the training data and an R2 value of 0.62, RMSE of 0.73, MSE of 0.53 and MAE of 0.55 for the test data, an indication of its reliability in making prediction. Cross-docking of cognate lorlatinib against ALK (4CLI) model yielded a binding mode closely aligned with the native conformer of lorlatinib, exhibiting an RMSD of 0.13 Å. Additionally, Induced Fit Docking (IFD) and Prime MM-GB/SA calculations indicated that briga_15 possesses the highest IFD Score of -675.27 kcal/mol, MM-GB/SA binding energy of -61.60 kcal/mol and a predicted pIC50 value of 8.57. The binding of briga_15 is facilitated by critical hydrogen bond networks with His1124 and Met1199 of ALK. Within the crizotinib series, crizo_35 demonstrated an exceptional IFD score of -671.16 kcal/mol, a predicted experimental pIC50 value of 7.75, and a binding energy of -77.71 kcal/mol, with water-mediated hydrogen bond networks involving Asp1203 and Lys1150 of ALK. These findings suggest that briga_15 and crizo_35 are promising leads warranting further optimization, synthesis, and biological evaluation.