This repo contains ongoing work exploring recurrent neural network models as applied to protein secondary structure prediction.
The pssp_lstm module contains an implementation of the LSTM RNN specified in Sonderby & Winther, 2015. See the README in that folder for more details and a user guide.
The lm_pretrain module allows users to train bidirectional language models that can be combined with bidirectional RNNs for protein secondary structure prediction. See the README in that folder for more details and a user guide.
This repo is under development. Current work is focusing on expanding the functionality of lm_pretrain to allow for more flexible models, and for exploring different ways of integrating pretrained LMs into BDRNNs.
Work on implementing models from other papers is currently on pause.