fw1121 / deephlapan

A deep learning approach for predicting high-confidence neoantigens by considering both the presentation possibilities of mutant peptides and the potential immunogenicity of pMHC

DeepHLApan

DeepHLApan is a deep learning approach used for predicting high-confidence neoantigens by considering both the presentation possibilities of mutant peptides and the potential immunogenicity of pMHC.

Contact: [e-mail]

Download and installation

There are two ways to install the DeepHLApan.

1.Docker (Recommend)

The Installation of Docker (v18.09) can be seen in https://docs.docker.com/

Pull the image of deephlapan from dockerhub:

  sudo docker pull biopharm/deephlapan:v1.1  

run the image in bash mode:

  sudo docker run -it --rm biopharm/deephlapan:v1.1 bash

2.Git (All the dependencies should be properly installed)

System

Linux

Dependencies

perl
python
cuda 9
cudnn 7

Steps

Download the latest version of DeepHLApan from https://github.com/jiujiezz/deephlapan

git clone https://github.com/jiujiezz/deephlapan.git

Unzip the source code and go into the directory by using the following command:

tar xvzf deephlapan-*.tar.gz

cd deephlapan

Invoke the setup script:

sudo python setup.py install

General usage

Single peptide and HLA:

deephlapan -P LNIMNKLNI -H HLA-A02:01 

List of peptides and HLA alleles in a file:

deephlapan -F [file] -O [output directory]  

Input files

DeepHLApan takes csv files as input with head of "Annotation,HLA,peptide" (requisite).
It supports to rank the HLA-peptide pairs if all the mutant peptides belong to one sample.

For example (demo/1.csv):

  Annotation,HLA,peptide
  NCI-3784,HLA-A01:01,MKRFVQWL
  NCI-3784,HLA-A03:01,MKRFVQWL
  NCI-3784,HLA-B07:02,MKRFVQWL
  NCI-3784,HLA-B07:02,MKRFVQWL
  NCI-3784,HLA-C07:02,MKRFVQWL
  NCI-3784,HLA-C07:02,MKRFVQWL
  NCI-3784,HLA-A01:01,KRFVQWLK
  NCI-3784,HLA-A03:01,KRFVQWLK
  NCI-3784,HLA-B07:02,KRFVQWLK
  NCI-3784,HLA-B07:02,KRFVQWLK
  NCI-3784,HLA-C07:02,KRFVQWLK
  NCI-3784,HLA-C07:02,KRFVQWLK

The content in Annotation can be changed as users wanted.

Update log

2019.12

V1.1.1
Improve the prediction speed

2019.03

V1.1
Add the function of immunogeneicity prediction

2018.07

V1.0
Test the suitabilty of different RNN variants (GRU,LSTM,BGRU,BLSTM,att-BGRU and att-BLSTM) on the binding prediction and select the best (att-BGRU) one for model construction.