hmMHC

A hidden Markov model-based MHC II binding predictor. Currently, only H2-IAb predictions are supported. The predictor is described in the paper

Elise Alspach et al. MHC-II neoantigens shape tumour immunity and response to immunotherapy. Nature (2019), available at https://www.nature.com/articles/s41586-019-1671-8

Installation

hmMHC can be installed on Linux and macOS via a combination of conda and pip install. Windows is not supported. Python 3 is not supported.

$ conda create -n hmmhc -c bioconda python=2.7 ghmm=0.9 'icu=58.*'
$ conda activate hmmhc
$ pip install git+https://github.com/artyomovlab/hmmhc#egg=hmmhc

Command line example

Input from command line and output to stout:

$ hmmhc-predict --allele H2-IAb --peptides VNGYNEAIVHVVETP IKSEHPGLSIGDVAK KESVVSGKAVPREEL

Input from csv and output to csv:

$ hmmhc-predict --allele H2-IAb --input example_input.csv --output output.csv

See hmmhc-predict -h for further details.

Python example

from hmmhc import hmMHC
predictor = hmMHC('H2-IAb')

peptides = ['VNGYNEAIVHVVETP', 'IKSEHPGLSIGDVAK', 'KESVVSGKAVPREEL']

predictor.predict(peptides)

Output

The predictor outputs a list of peptides with the predicted -10 log odds scores and corresponding percentile ranks. Percentile ranks are computed from -10 log odds scores based on model calibration on a large set of random natural peptides. For both metrics, smaller values correspond to higher binding likelihood. See Methods section in the paper for further details.

Dependencies

hmMHC relies on General Hidden Markov Model library (GHMM) by A. Schliep et al., see http://ghmm.sourceforge.net/.

Latest version

The latest version of hmMHC is available at https://github.com/artyomovlab/hmmhc.