PresentER-ML

Exploratory data anlysis of and training of a prediction tool for TCR off-target peptides from PresentER screen data.

Requirements

• Python libraries: pandas, biopython, seaborn, numpy, mhctools, sklearn
• Locally installed NetMHC4, MixMHCpred

Manifest

assemble_df.py: Reformats data from an investigator performing a PresentER screen in donor TCR-transduced T cells

physicochemical_annotation.py: Annotates each peptide with physicochemical properties (e.g. hydrophobicity, molecular weight, isoelectric point). Produces example plot of PresentER depletion score (diff) vs. molecular weight (mw) (peptide sequences redacted).

binding_preds.py: computes HLA-A*02:01 predicted binding for screened peptides using different MHC binding predictors physicochemical_preds.py: Combines physicochemical annotations with binding prediction annotations. pairplot.py: Visualization of peptide features in annotatated dataset that could principally underlie off-target character in PresentER screen.

pca.py: Produces PCA based on annotated dataframe.

train_nb.py: Train naive bayes classifier to predict if a given peptide will be depleted or not based on 2-grams