The function of most proteins is intricately linked to the complex network of interactions they make. It is not surprising, therefore, that these interfaces between protein-protein interactions (PPIs) are enriched in disease mutations and the focus of drug development efforts. Understanding how mutations affect PPIs is vital for understanding their role in diseases, pathogenicity, drug resistance and to guide rational engineering. While several prediction tools have been advanced, a prevalent issue has been their compromise between speed and accuracy. In response to this challenge, we present DDMut-PPI, a deep learning model that quickly and accurately predicts how missense mutations change protein-protein binding free energy. Building upon the robust siamese network architecture with graph-based signatures from our prior work, DDMut, the DDMut-PPI model was further augmented with a graph convolutional network (GCN) operated on the protein interaction interface. We employed the cutting-edge ESM-2 protein language model to provide residue-specific embeddings as node features within the graph. Additionally, a variety of molecular interactions served as edge features, ensuring that the graph neural network representation provides a comprehensive depiction of interface interactions. By integrating evolutionary context with spatial information, this sophisticated framework enables DDMut-PPI to detect both proximate and distal effects of mutations, yielding a robust Pearson's correlation of up to 0.70 (RMSE: 1.46 Kcal/mol) in our evaluations, outperforming most existing methods. Importantly, the model demonstrated consistent performance across mutations, irrespective of their propensity to strengthen or weaken PPIs. DDMut-PPI offers a significant advancement in the field and will serve as a valuable tool for researchers probing the complexities of protein interactions. DDMut-PPI is freely available as a web server at https://biosig.lab.uq.edu.au/ddmut_ppi.