def make_prediction(peptide, allele_sequence, model=None):
mhc_seq = padded_indices([allele_sequence],
add_start_symbol=False,
add_end_symbol=False,
index_dict=amino_acid_letter_indices)
#returns an array of index encoded peptide/peptides depending on peptide length
X_p = padded_indices(format_peptide(peptide),
add_start_symbol=False,
add_end_symbol=False,
index_dict=amino_acid_letter_indices)
#tiling the mhc in case the peptide is more than a length of 9
mhc_seq = np.tile(mhc_seq,(len(X_p),1))
preds = 0
#mean of the predicted outputs in case peptide is more than length of 9
if(model):
preds = model.predict({'peptide':X_p,'mhc':mhc_seq})['output']
preds = np.mean(preds)
return preds
def get_model_data(allele_list,
allele_sequence_data,
allele_binding_data,
peptide_length =9,
mhc_length=None,
):
'''
generate training data for a list of alleles where
the output is index_encoded peptide, dense MHC sequence
and the log-transformed ic50 values
'''
data_len =0
for allele in allele_list:
data_len += len(allele_binding_data[allele][1])
X_p = np.zeros((data_len,peptide_length))
X_mhc = np.zeros((data_len,mhc_length))
Y_combined = np.zeros(data_len)
index=0
for allele in allele_list:
peptides = allele_binding_data[allele][1]
end_index = index+len(peptides)
#index encoding for peptides
X_p[index:end_index] = padded_indices(peptides,
add_start_symbol=False,
add_end_symbol=False,
index_dict=amino_acid_letter_indices)
#dense vector for mhc
mhc_seq = padded_indices([allele_sequence_data[allele]],
add_start_symbol=False,
add_end_symbol=False,
index_dict=amino_acid_letter_indices)
X_mhc[index:end_index] = np.tile(mhc_seq,(len(peptides),1))
#log-transformed binding values
Y_combined[index:end_index] = allele_binding_data[allele][0]
index+=len(peptides)
arr = np.arange(len(X_p))
np.random.shuffle(arr)
X_p_shuffled = X_p[arr]
X_mhc_shuffled = X_mhc[arr]
log_binding_values_shuffled = Y_combined[arr]
return X_p_shuffled, X_mhc_shuffled, log_binding_values_shuffled
