To customize treatment based on an individual patient’s unique characteristics, personalized medicine relies on predicting drug response. Such a method maximizes benefits and minimizes side effects. It simplifies choosing which therapy to use by reducing trial-and-error approaches and saves time as well. This also effectively improves patient outcomes while saving costs through needless medical treatment avoidance. This study focuses on the performance in terms of evaluation metrices such as training loss, and F1-score of different graph-based models on drug response prediction when they are combined with different feature extraction methods. In this study, five different graph models are used which are the Graph Convolutional Network, Graph Attention Network, GraphSAGE model, Graph Isomorphism Network, and Graph Transformers, and two feature extraction methods, CNN and LSTM. The investigation ultimately predicts response values for all drug-cell line pairs using a densely connected neural network and compares the performance of different graph models.