The long and boring (but academically correct) abstract is available at the publisher (read Abstract at Frontiers). Here we will provide a cooler version of what the paper is about. APRANK is an Antigenic Protein (and peptide) Ranker. To develop APRANK we used information on known and validated antigens and their peptidic epitopes (spoiler alert: this is all biased towards linear epitopes), and used it to train two models (one for proteins, to predict antigens; one for peptides to predict epitopes). APRANK takes a complete proteome as input, calculates a number of features for each protein and peptide, and outputs a score of how likely it is that the given protein or peptide is antigenic. APRANK has been trained with antigenic proteins and peptides from a phylogenetically wide diversity of human pathogens, and has been validated to works for bacteria, protozoa, and metazoan parasites (helminths).
APRANK is available to predict antigenicity but also for hacking and modifying.
We have made all code, data and trained models available. Code is under a simplified BSD 2-clause license, which is a permisive free software license. Feel free to reuse, retrain, and take APRANK to the next level.
If you read the paper, you will get pointers to all the relevant places, but in short:
Hopefully everything is well documented in the code! Enjoy.