Predicting PKB/Akt Downstream Signaling Blockers Using Machine Learning Models

Cancer is one of the root causes of unusual deaths in the world. To prevent cancer, Protein Kinase B (PKB) is one of the targets utilized to discover anticancer drugs. The quantitative structure-activity relationship method has been extended with machine learning approaches to select drug candidates for inhibiting PKB/Akt downstream signaling cascades. The study incorporated 76 probable PKB/Akt signaling blockers from the ChEMBL database. After preprocessing to handle the entries with missing bioactivity data, the compounds are distinguished into active and inactive categories based on their IC50 values. The research includes the calculation of molecular descriptors of Lipinksi’s rule of five for analyzing the absorption, distribution, metabolism, and excretion properties of the small ligand molecules to analyze their capability to act like active drugs. The bioactivity data is computed by PaDel descriptor to enumerate the PubChem fingerprints in binary patterns. Out of 34 machine learning regression predictive models that have been used to train on fingerprint data to predict the drug-likeness of the small molecules, the best-performing four are XGBoost, DecisionTree, GaussianProcess, and ExtraTree. These models have shown good fits to the data demonstrating coefficients of determination values: 0.9 for each and high potential to predict the drug ligands. This study establishes a novel approach for anticancer drug discovery to restore the regulation of abnormal PI3K/Akt signaling pathways.