Protein Secondary Structure Prediction

Reference implementation for the convolutional–recurrent Q8 sequence model described in the accompanying research report.

Files

• main.py — training, recorded validation runs, model export, and prediction export
• model.py — network classes and model construction
• data_prep.py — tokenization, encoding, padding, and target preparation
• util.py — n-gram conversion, masked accuracy, and prediction decoding

Environment note

The code uses the Keras API available when the model was developed. Current TensorFlow/Keras releases may require import and optimizer-argument updates. The source is provided as a reference implementation rather than a pinned, production-ready package.

Data

The training and test datasets are not included in this download. The scripts expect dataset/train.csv and dataset/test.csv with the columns shown in main.py.