Machine learning-guided drug repurposing for targeting PDK1 in breast cancer treatment

Abstract

The phosphoinositide 3-kinase (PI3K)/3-phosphoinositide-dependent protein kinase-1 (PDK1)/protein kinase B (AKT) signaling axis is dysregulated in breast cancer progression and therapeutic resistance. In this study, an integrated computational workflow was developed to prioritize FDA-approved drugs as potential PDK1-targeting candidates using machine learning-based quantitative structure-activity relationship (QSAR) modeling, applicability-domain (AD) assessment, molecular docking, molecular dynamics (MD) simulation, and absorption, distribution, metabolism, excretion, and toxicity (ADMET) prediction. We used a curated set of 1,971 compounds with experimentally reported PDK1 inhibitory activity to develop five regression models. The Random Forest model performed best with 10-fold cross-validation R² = 0.83, RMSE = 0.60, MAE = 0.44, and an independent test set based on scaffolds R² = 0.84, RMSE = 0.58, MAE = 0.44. Among 1,615 structurally eligible FDA-approved compounds, 83 compounds had predicted IC₅₀ values below 500 nM. Allopurinol was the only FDA-screened compound accepted by all six complementary AD methods in the consensus applicability-domain analysis. The prioritized compounds were docked at the ATP-binding site of PDK1, and the predicted docking score ranged from − 10.6 to -6.5 kcal/mol, with Allopurinol having the best score. We used comparative 100 ns MD simulations to prioritize Allopurinol for further analysis. A subsequent 1,000 ns simulation confirmed the persistent association with the ATP-binding region, but with dynamic rearrangement; the overall compactness of the protein was maintained. We analyzed overall structural behavior and conformational changes using trajectory analysis based on radius of gyration (Rg), solvent-accessible surface area (SASA), principal component analysis (PCA), and free-energy landscape (FEL). The ADMET properties were generally good, with predictions of drug-likeness and permeability, and potential pharmacokinetic and toxicity liabilities that need experimental evaluation. Overall, Allopurinol was computationally prioritized for further biochemical, cellular, and in vivo validation.