Sequence-based protein class prediction.

Dependencies

• python 3.11.0
• scikit-learn 1.4.2
• h5py 3.11.0
• pytorch 1.0.2
• position_encoding (github.com/cmbi/position-encoding)

Data Format

The input data must be presented as an hdf5 file, containing the following:

• peptide/sequence_onehot: a 9x22 array, containing one-hot encoded amino acid sequence
• affinity: a value between 0.0 (low affinity) to 1.0 (high affinity)

Training & testing a flattening model

python run.py flattening train.hdf5 valid.hdf5 test.hdf5 results.csv