def merge_seq_with_mpra(dataset):
''' converts DNA sequences around variants into dataframe, merges with MPRA
train and test sets '''
train = load_train_set(dataset)
test = load_test_set(dataset)
seq_df = convert_seq_fa_to_df(os.path.join(cfg.SEQ_DIR, SeqDatasets))
seq_df['pos'] = seq_df['pos1'] + 9
train_seq = pd.merge(train[['chr','pos','rs','Label']], seq_df[['chr','pos','seq']], how='inner')
test_seq = pd.merge(test[['chr','pos','rs','Label']], seq_df[['chr','pos','seq']], how='inner')
return train_seq, test_seq
def add_valley_scores(dataset):
dataset='E116'
train = load_train_set(dataset)
test = load_test_set(dataset)
valley = pd.read_csv(os.path.join(cfg.DATA_DIR, 'bigwig', '{0}_valley.csv'.format(dataset)))
cols = valley.columns.values
new_cols = ['val' + x for x in cols]
new_cols[0] = 'variant'
valley.columns = new_cols
val_train = pd.merge(train[['chr','pos','rs','Label']], valley, left_on='rs', right_on='variant')
val_test = pd.merge(test[['chr','pos','rs','Label']], valley, left_on='rs', right_on='variant')
return val_train, val_test
'''
all_results = pd.DataFrame()
param_storage = []
random_param_list = [param_fn() for _ in range(iterations)]
for iter in range(iterations):
print('Iteration {0}: '.format(iter))
params = random_param_list[iter]
trials = splits_single_model(data, params, mod='mpra')
params.update(trials)
param_storage.append(params)
# periodically save results to file and flush param_storage
if iter % 2 == 1:
all_results = pd.concat([all_results, pd.DataFrame(param_storage)])
all_results.to_csv(outfile, index=False)
param_storage = []
if __name__ == '__main__':
data = load_train_set(dataset='E116')
outfile = join(cfg.OUTPUT_DIR, 'hparams_vat2.csv')
execute_random_search(data,
outfile=outfile,
iterations=50,
param_fn=get_random_params_vat)
fig.savefig(join(cfg.OUTPUT_DIR, 'f1.png'), dpi=150, bbox_inches='tight')
tsne2 = sns.lmplot(x='tsne-1',
y='tsne-2',
data=df_tsne,
hue='Predicted',
fit_reg=False,
palette='Set2')
plt.title('t-SNE components by predicted label, threshold=0.03')
fig = tsne2.fig
fig.savefig(join(cfg.OUTPUT_DIR, 'f2.png'), dpi=150, bbox_inches='tight')
# error analysis
res = pd.read_csv(join(cfg.OUTPUT_DIR, 'scores.csv'))
train = load_train_set('E116')
train_counts = train.groupby('chr').sum().Label
errs = np.zeros(22)
for c in range(1, 23):
tmp = res[res['chr'] == c]
met = avgPR(tmp['Label'], tmp['Score'])
errs[c - 1] = met
chrs = np.arange(1, 23)
col1 = '#66c2a5'
col2 = '#fc8d62'
fig, axes = plt.subplots(nrows=2, sharex=True)
axes[0].bar(chrs, errs, color=col1)
axes[1].bar(chrs, train_counts, color=col2)