allergen-net allows to predict if a protein sequence is an allergen. The project is created as part of the annual iGEM competition 2019 by the team of the Friedrich-Alexander-University Erlangen-Nuremberg.

python train.py --timestamp "20190807-102903" --filters=128 --depth=4 --epochs 150

The --timestamp option determines which datasplit to use (standard split is 1, but one can switch to another one via --split). Normally the network trains from scratch, unless a --model=<FILENAME> option is supplied, then it continues training with the checkpoint.

python analyse.py --test <path-to-test-set> --model <path-to-network> --acc --loss --roc --pr

Depending on the options, the history, the receiver operator characteristics and the precision recall curve are plotted to a file in the current directory. If --output=<path> is supplied, then the plots are stored in the given directory.