I am fairly new to neural networks, I have experience implementing a few simple CNN and RNNs. I really appreciate any advice.

I am looking to train a NN to predict the binding affinity for protein targets from chemical structure. I have a set of molecules and their binding affinities.

I was thinking of representing the molecules in the SMILES format. For example, popionoic acid is "CCC(=O)O". Say I am concerned with 3 binding targets, and this molecule has the affinities 1.5, 2.1, and 3.2. I would like to train my network to predict these affinities from the structure specified as a SMILES string.

One major problem I am facing is the varying input size of different molecules. I know RNN are good for handling this, but I am unsure how to start designing the network. I am unsure how to design a RNN to predict multiple affinity values given a chemical structure.

Is there a way I could normalize the inputs to fit a traditional CNN?

Does anyone know of implemented networks modeling a similar problem?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.