def get_uncorr_drug_features(include_rnaseq=True):
# first get the compounds that are from the LINCS dataset
drug_features_dict = get_feature_dict('Data/phase2_compounds_morgan_2048.csv') # , use_int=True)
drug_features_dict = remove_non_lncap(drug_features_dict)
# add the compounds that were in RNAseq
if include_rnaseq:
rnaseq_drugs = get_feature_dict('Data/inhouse_morgan_2048.csv')
for rnaseq_drug in rnaseq_drugs:
drug_features_dict[rnaseq_drug] = rnaseq_drugs[rnaseq_drug]
unique_drug_features_dict = remove_dups(drug_features_dict)
drug_features = get_array(unique_drug_features_dict)
no_cor_np = remove_corr_features(drug_features)
return remove_dup_np(no_cor_np)
def get_predictions(up_model_filename, down_model_filename):
# load the models
up_model = load_model_from_file_prefix(up_model_filename)
down_model = load_model_from_file_prefix(down_model_filename)
# build your input features
gene_features_dict = get_feature_dict("Data/go_fingerprints.csv")
drug_features_dict = get_feature_dict("Data/inhouse_morgan_2048.csv")
#drug_features_dict.pop("Enzalutamide")
#drug_features_dict.pop("VPC14449")
# drug_features_dict.pop("VPC17005")
data = []
descriptions = []
rnaseq_missing_genes = [ # these genes were not in the rnaseq dataset
'GATA3',
'RPL39L',
'IKZF1',
'CXCL2',
'HMGA2',
'TLR4',
'SPP1',
'MEF2C',
'PRKCQ',
'MMP1',
'PTGS2',
'ICAM3',
'INPP1',
]
for gene in rnaseq_missing_genes:
gene_features_dict.pop(gene, None)
for drug in drug_features_dict:
for gene in gene_features_dict:
data.append(drug_features_dict[drug] + gene_features_dict[gene])
descriptions.append(drug + ", " + gene)
data = np.asarray(data, dtype=np.float16)
# get predictions
up_predictions = up_model.predict(data)
down_predictions = down_model.predict(data)
return up_predictions, down_predictions, drug_features_dict, gene_features_dict
def get_jaccard_score_of_rnaseq_drug(drug_id, lincs_drugs):
rnaseq_drugs = get_feature_dict('Data/inhouse_morgan_2048.csv')
rnaseq_drug = rnaseq_drugs[drug_id]
rnaseq_drug = np.reshape(np.array(rnaseq_drug, np.float16), (1, -1))
rnaseq_drug = remove_corr_features(rnaseq_drug)
scores = []
for lincs_drug in lincs_drugs:
score = jaccard_similarity_score(lincs_drug, rnaseq_drug[0])
scores.append(score)
return np.mean(scores)
return 0
else:
return highest_vote_class
def get_their_id(good_id):
return 'b\'' + good_id + '\''
def get_our_id(bad_id):
return bad_id[2:-1]
# get the dictionaries
print(datetime.datetime.now(), "Loading drug and gene features")
drug_features_dict = get_feature_dict('Data/phase2_compounds_morgan_2048.csv')
gene_features_dict = get_feature_dict('Data/go_fingerprints.csv')
cell_name_to_id_dict = get_feature_dict('Data/Phase2_Cell_Line_Metadata.txt',
'\t', 2)
experiments_dose_dict = get_feature_dict(
LINCS_data_path + 'GSE70138_Broad_LINCS_sig_info.txt', '\t', 0)
gene_id_dict = get_gene_id_dict()
lm_gene_entrez_ids = []
for gene in gene_id_dict:
lm_gene_entrez_ids.append(get_their_id(gene))
print("Loading gene expressions from gctx")
level_5_gctoo = load_gene_expression_data(
LINCS_data_path + "GSE70138_Broad_LINCS_Level5_COMPZ_n118050x12328.gctx",
lm_gene_entrez_ids)