DeepPeptide

Predicting cleaved peptides in protein sequences.

Training the model

1. Precompute embeddings using src/utils/make_embeddings.py
2. Train the model

python3 run.py --embeddings_dir PATH/TO/EMBEDDINGS -df data/labeled_sequences.csv -pf data/graphpart_assignments.csv

Note that parameters --lr, --batch_size, --dropout, --conv_dropout, --kernel_size, --num_filters, --hidden_size were optimized in a nested CV hyperparameter search and not used at their defaults.

Evaluation

• PeptideLocator was evaluated as a licensed executable and cannot be provided in this repo.
• We used 5-fold nested CV to select 20 model checkpoints trained using src/train_loop_crf.py. The selected checkpoints are hardcoded in evaluation/measure_performance.py, which computes the performance metrics from the checkpoints' saved predictions.

Predicting

See the predictor README