Question on using real biological network as artifical neural network architecture

I am trying to use the topology of a real life cell signaling pathway as the architecture in an ANN and then train the model on bioinformatics data. R is my tool of choice, but I can't find any R package that allows me to actually provide the network structure as an argument to the neural network fitting function, they only let you choose from commonly used models. The models I am using are typically feed forward, but in some cases have cycles. The constraint is the R packages are not flexible enough. I am not new to machine learning, but I am to neural networks and deep learning and all that. What software tools allow you to pick your own topology?