Predicting associations between drugs and adverse side effects is essential for drug discovery and safety evaluation. Current models predominantly emphasize singular attributes of drugs and side effects, often neglecting to fully encapsulate their multifaceted characteristics and intricate interrelations. Here, we present MFGNN-DSA, a multifeature graph neural network framework that integrates heterogeneous biomedical information to achieve a more accurate prediction. Initially, the model extracts multisource features of drugs and side effects, which are integrated into attribute-based feature vectors via graph sampling and aggregation networks. A heterogeneous network encompassing diseases, drugs, and side effects is then constructed, and the HIN2Vec method is applied to obtain topological feature vectors. Subsequently, these topological, attribute-based, and aggregated feature vectors are processed through a multihead self-attention mechanism to derive the final feature vectors. Ultimately, the concatenated feature vectors are passed through a fully connected layer to predict the probability of drug-side effect association. Experimental results demonstrate that our model outperforms state-of-the-art methods in terms of AUC and AUPR. Case studies offer additional evidence supporting the model's effectiveness. The source code and experimental data of MFGNN-DSA are publicly available at https://github.com/MFGNN/MFGNN-DSA.