How to construct samples that contain both graphs and other features? For example, (graph, feature_1, feature_2, ...).

[arnoegw]

The input to a Keras Model can be any nest of tensors, including the kind of tuple you sketched.

For deserialization from tf.train.Example, esp. in the presence of more than one GraphTensor, the prefix= argument of tfgnn.parse_example() might come in handy.

Alternatively, if each example has one graph plus some freestanding features, consider packing the freestanding features into the Graph.context, see the Data Preparation and Sampling guide.