Hypertension is a major global cardiovascular disease, with ACE inhibitors central to its treatment. Accurate prediction of compound bioactivity is essential to accelerate drug discovery. This study proposes a regression model using Long Short-Term Memory (LSTM) networks optimized via Grey Wolf Optimization (GWO) to predict the inhibitory activity (pIC50) of ACE-targeting compounds. The model utilizes SMILES-based molecular representations encoded into vector sequences, eliminating the need for handcrafted descriptors. Although the dataset contains only 255 bioactive compounds from ChEMBL, prior research shows that small datasets can yield meaningful insights with proper modeling. Sequential learning from SMILES enables the model to capture temporal patterns and mitigate data limitations. Several LSTM baselines were evaluated, followed by GWO-based hyperparameter tuning. The best model achieved a test (R2) of 0.667 comparable to prior Random Forest (R2 ? 0.745) and SVM (R2 ? 0.74) results. Additionally, the model obtained a leave-one-out cross-validated coefficient of determination (Q2LOO) of 0.696, indicating strong internal consistency. These results suggest that combining sequence-based modeling with metaheuristic optimization provides a robust and efficient method for predicting ACE inhibitor bioactivity.