Is it possible to train a layer's weights but not its biases (or vice versa) in Keras?

Question

I know from this question that it is possible to set an entirely layer as non-trainable by something like this:

frozen_layer = Dense(32, trainable=False)