MHC binding prediction based on modeled physicochemical properties of peptides
Project description
MHCLovac
MHC binding prediction based on modeled physicochemical properties of peptides.
New release - MHCLovac 2.0
MHCLovac 2.0 makes almost somewhat accurate predictions. It really is not precise.
About
MHCLovac uses Bayesian linear regression for binding affinity prediction
based on modeled physicochemical properties of peptides.
MHCLovac uses pre-developed proteinko
package to obtain modeled distributions of physicochemical properties.
Physicochemical properties in question are:
- Hydropathy
- Number of donor hydrogen bonds
- Number of acceptor hydrogen bonds
- Isoelectric point
- Van der Waals molecular volume
Once the distributions are obtained, the area under the curve (AUC) is calculated using a sliding frame technique. The AUC values for each of five physicochemical properties are concatenated into single feature vector.
Model training is performed on standardized AUC values. We tested number of
linear regression models and concluded that BayesianRidge algorithm from sklearn
package produces most consistent predictions across various training set
configurations.
MHCLovac makes modestly accurate predictions, which can be seen on plots below.
Installation
Install from PyPI repository
pip install mhclovac
Download and install from git repository
git clone https://github.com/stefs304/mhclovac
cd mhclovac
pip install .
Usage
mhclovac --fasta <fasta file>
--hla <hla type (ex. HLA-A*02:01)>
--peptide_length <peptide length>
--output <output file (optional)>
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.