EpiDope creation space

This repo contains most of the code used to create https://github.com/flomock/EpiDope.
See also our paper https://doi.org/10.1093/bioinformatics/btaa773.

We have numerous scripts with specific niche functionality. Therefore, we expect that most of the scripts are not of high interest for most users. Because of this, we only rudimentarily polished most of the code and do not guaranty it's functionality.

possible usage

get the raw data

download from http://www.iedb.org/bcelldetails_v3.php
export to csv file

download positive and negative samples

utils/download_from_iedb.py
(change local variables like line 17 (path to csv file))

utils/download_proteins_from_epitopeNumber.py (change local variables like line 13 (path to csv file))

curate data

curate_iedb_linear_epitopes.py
(again changing input path)

make cluster by different sequence identity:

cd previous_output_dir
cat * >> protein_all.fasta

cd-hit -i protein_all.fasta -c 1 -o 1_seqID.fasta
cd-hit -i protein_all.fasta -c 0.9 -o 0.9_seqID.fasta
cd-hit -i protein_all.fasta -c 0.8 -o 0.8_seqID.fasta
cd-hit -i protein_all.fasta -c 0.7 -o 0.7_seqID.fasta
cd-hit -i protein_all.fasta -n 4 -c 0.6 -o 0.6_seqID.fasta
cd-hit -i protein_all.fasta -n 3 -c 0.5 -o 0.5_seqID.fasta

select proteins with most verified regions
utils/cluster_to_proteins_with_markings.py

generate training, test, val set

simple clustered
generate_binary_clustered_training_sets.py

more complex, if your data is clustered twice (like in the paper explained), to reduce bias of similar sequences in the test set.
generate_binary_double_clustered_training_sets.py

train and test the models

train_DL.py trains multiple neural networks on your training data
epidope.py testing suit for trained your models

further

make multi fasta file with only test set entries
utils/filter_test_set_fastas.py

get the ROC precision-recall und distribution of predictions:
utils/make_ROC_curves.py

get plots with only the parts marked which are part of ROC
utils/plots_test_set.py