Smarter Pooling Mechansims

Hi, I am currently using global max pooling and global mean pooling for a Graph Convolutional Network using PyTorch Geometric. Since my input data is in a one-hot encoded format I believe such pooling mechanisms are not sufficient to hold enough information to train my model. Are there better pooling mechanisms suitable for one-hot encoded format data? As a beginner, any suggestions or guidance will be really appreciated.