def calc_statistics(orig_df1, orig_df2, exp, alldf_dict):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df1 = _lib.mh_del_subset(orig_df1)
df1 = _lib.indels_without_mismatches_subset(df1)
df2 = _lib.mh_del_subset(orig_df2)
df2 = _lib.indels_without_mismatches_subset(df2)
if bool(sum(df1['Count']) <= 1000) or bool(sum(df2['Count']) <= 1000):
return
df1['Frequency'] = _lib.normalize_frequency(df1)
df2['Frequency'] = _lib.normalize_frequency(df2)
join_cols = ['Category', 'Genotype Position', 'Length']
mdf = df1.merge(df2, how='outer', on=join_cols, suffixes=['_1', '_2'])
mdf['Frequency_1'].fillna(value=0, inplace=True)
mdf['Frequency_2'].fillna(value=0, inplace=True)
r = pearsonr(mdf['Frequency_1'], mdf['Frequency_2'])[0]
print exp, r
alldf_dict['_Experiment'].append(exp)
return alldf_dict
def calc_statistics(orig_df, exp, alldf_dict):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
df = df[df['Length'] >= 5]
if sum(df['Count']) <= 500:
return
df['Frequency'] = _lib.normalize_frequency(df)
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(df)
pred_df = _predict2.predict_mhdel_cpf1(seq, cutsite)
join_cols = ['Category', 'Genotype Position', 'Length']
mdf = df.merge(pred_df, how='outer', on=join_cols)
mdf['Frequency'].fillna(value=0, inplace=True)
mdf['Predicted_Frequency'].fillna(value=0, inplace=True)
obs = mdf['Frequency']
pred = mdf['Predicted_Frequency']
r = pearsonr(obs, pred)[0]
alldf_dict['gt_r'].append(r)
alldf_dict['_Experiment'].append(exp)
return alldf_dict
def calc_statistics(df, exp, alldf_dict):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(df)
df = _lib.indels_without_mismatches_subset(df)
df = df[df['Length'] >= 5]
if sum(df['Count']) <= 1000:
return
df['Frequency'] = _lib.normalize_frequency(df)
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(df)
pred_df = _predict2.predict_mhdel(seq, cutsite)
pred_df = pred_df[pred_df['Length'] >= 5]
pred_df['Predicted_Frequency'] = pred_df['Predicted_Frequency'] / sum(
pred_df['Predicted_Frequency'])
join_cols = ['Category', 'Genotype Position', 'Length']
mdf = df.merge(pred_df, how='outer', on=join_cols)
mdf['Frequency'].fillna(value=0, inplace=True)
mdf['Predicted_Frequency'].fillna(value=0, inplace=True)
obs = mdf['Frequency']
pred = mdf['Predicted_Frequency']
ns_criteria = (mdf['Length'] - mdf['Genotype Position'] == 0)
s = mdf[ns_criteria]
alldf_dict['Predicted Ngt'] += list(s['Predicted_Frequency'])
alldf_dict['Observed Ngt'] += list(s['Frequency'])
alldf_dict['_Experiment'] += [exp] * len(s['Frequency'])
return alldf_dict
def featurize(orig_df):
seq, cutsite = _lib.get_sequence_cutsite(orig_df)
mh_lens, gc_fracs, del_lens, freqs = [], [], [], []
dl_freqs = []
DELLEN_LIMIT = 60
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
df = df[df['Length'] <= DELLEN_LIMIT]
if sum(df['Count']) < 1000:
return None
criteria = (orig_df['Category'] == 'del') & (orig_df['Length'] <= 28)
s = orig_df[criteria]
s['Frequency'] = _lib.normalize_frequency(s)
for del_len in range(1, 28 + 1):
dl_freq = sum(s[s['Length'] == del_len]['Frequency'])
dl_freqs.append(dl_freq)
df['Frequency'] = _lib.normalize_frequency(df)
for del_len in range(1, DELLEN_LIMIT + 1):
left = seq[cutsite - del_len:cutsite]
right = seq[cutsite:cutsite + del_len]
mhs = find_microhomologies(left, right)
for mh in mhs:
mh_len = len(mh) - 1
if mh_len > 0:
gtpos = max(mh)
s = cutsite - del_len + gtpos - mh_len
e = s + mh_len
mh_seq = seq[s:e]
gc_frac = get_gc_frac(mh_seq)
criteria = (df['Length'] == del_len) & (df['Genotype Position']
== gtpos)
freq = sum(df[criteria]['Frequency'])
mh_lens.append(mh_len)
gc_fracs.append(gc_frac)
del_lens.append(del_len)
freqs.append(freq)
return mh_lens, gc_fracs, del_lens, freqs, dl_freqs
def calc_statistics(orig_df, exp, alldf_dict):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
if sum(df['Count']) <= 1000:
return
df['Frequency'] = _lib.normalize_frequency(df)
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(df)
pred_df = _predict2.predict_mhdel(seq, cutsite)
join_cols = ['Category', 'Genotype Position', 'Length']
mdf = df.merge(pred_df, how = 'outer', on = join_cols)
mdf['Frequency'].fillna(value = 0, inplace = True)
mdf['Predicted_Frequency'].fillna(value = 0, inplace = True)
obs = mdf['Frequency']
pred = mdf['Predicted_Frequency']
obs_entropy = entropy(obs) / np.log(len(obs))
pred_entropy = entropy(pred) / np.log(len(pred))
alldf_dict['obs gt entropy'].append(obs_entropy)
alldf_dict['pred gt entropy'].append(pred_entropy)
df = orig_df[orig_df['Category'] == 'del']
df = df[df['Length'] <= 28]
df['Frequency'] = _lib.normalize_frequency(df)
obs_dl = []
for del_len in range(1, 28+1):
freq = sum(df[df['Length'] == del_len]['Frequency'])
obs_dl.append(freq)
pred_dl = _predict2.deletion_length_distribution(seq, cutsite)
obs_entropy = entropy(obs_dl) / np.log(len(obs_dl))
pred_entropy = entropy(pred_dl) / np.log(len(pred_dl))
alldf_dict['obs dl entropy'].append(obs_entropy)
alldf_dict['pred dl entropy'].append(pred_entropy)
alldf_dict['_Experiment'].append(exp)
return alldf_dict
def calc_statistics(orig_df, exp, rate_model, bp_model, alldf_dict, rs, data_nm):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
if sum(df['Count']) <= 1000:
return
obs_d = defaultdict(list)
df = orig_df
# Grab observed deletions, MH and MH-less
for del_len in range(1, 59+1):
crit = (df['Category'] == 'del') & (df['Indel with Mismatches'] != 'yes') & (df['Length'] == del_len)
s = df[crit]
mh_s = s[s['Microhomology-Based'] == 'yes']
for idx, row in mh_s.iterrows():
obs_d['Count'].append(row['Count'])
obs_d['Genotype Position'].append(row['Genotype Position'])
obs_d['Length'].append(row['Length'])
obs_d['Category'].append('del')
mhless_s = s[s['Microhomology-Based'] != 'yes']
obs_d['Length'].append(del_len)
obs_d['Count'].append(sum(mhless_s['Count']))
obs_d['Genotype Position'].append('e')
obs_d['Category'].append('del')
obs_df = pd.DataFrame(obs_d)
# Grab observed 1 bp insertions
ins_crit = (orig_df['Category'] == 'ins') & (orig_df['Length'] == 1) & (orig_df['Indel with Mismatches'] != 'yes')
ins_df = orig_df[ins_crit]
truncated_ins_d = defaultdict(list)
for ins_base in list('ACGT'):
crit = (ins_df['Inserted Bases'] == ins_base)
tot_count = sum(ins_df[crit]['Count'])
truncated_ins_d['Count'].append(tot_count)
truncated_ins_d['Inserted Bases'].append(ins_base)
truncated_ins_d['Category'].append('ins')
truncated_ins_d['Length'].append(1)
ins_df = pd.DataFrame(truncated_ins_d)
obs_df = obs_df.append(ins_df, ignore_index = True)
obs_df['Frequency'] = _lib.normalize_frequency(obs_df)
crispr_subset = _lib.crispr_subset(orig_df)
frac_explained = sum(obs_df['Count']) / sum(crispr_subset['Count'])
# print frac_explained
# Save this for aggregate plotting
alldf_dict['Fraction Explained'].append(frac_explained)
# Predict MH dels and MH-less dels
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(orig_df)
pred_df = _predict2.predict_indels(seq, cutsite,
rate_model, bp_model)
mdf = obs_df.merge(pred_df, how = 'outer', on = ['Category', 'Genotype Position', 'Inserted Bases', 'Length'])
mdf['Frequency'].fillna(value = 0, inplace = True)
mdf['Predicted_Frequency'].fillna(value = 0, inplace = True)
r = pearsonr(mdf['Frequency'], mdf['Predicted_Frequency'])[0]
# Merge observed and predicted, double check correlation
# Store in a way that I can plot it.
data_nm_out_dir = out_dir + data_nm + '/'
util.ensure_dir_exists(data_nm_out_dir)
exp_out_dir = data_nm_out_dir + exp + '/'
util.ensure_dir_exists(exp_out_dir)
out_fn = exp_out_dir + '%.3f.csv' % (r)
mdf.to_csv(out_fn)
# Store in alldf_dict
alldf_dict['_Experiment'].append(exp)
alldf_dict['rs'].append(rs)
return alldf_dict
def calc_statistics(orig_df, exp, rate_model, bp_model, alldf_dict, rs):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
if sum(df['Count']) <= 1000:
return
ins_crit = (orig_df['Category'] == 'ins') & (orig_df['Length'] == 1)
ins_df = orig_df[ins_crit]
truncated_ins_d = defaultdict(list)
for ins_base in list('ACGT'):
crit = (ins_df['Inserted Bases'] == ins_base)
tot_count = sum(ins_df[crit]['Count'])
truncated_ins_d['Count'].append(tot_count)
truncated_ins_d['Inserted Bases'].append(ins_base)
truncated_ins_d['Category'].append('ins')
truncated_ins_d['Length'].append(1)
ins_df = pd.DataFrame(truncated_ins_d)
df = df.append(ins_df, ignore_index=True)
df['Frequency'] = _lib.normalize_frequency(df)
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(orig_df)
pred_df = _predict2.predict_mhdel(seq, cutsite)
# Predict rate of 1 bp insertions
# Featurize first
del_score = _predict2.total_deletion_score(seq, cutsite)
dlpred = _predict2.deletion_length_distribution(seq, cutsite)
norm_entropy = entropy(dlpred) / np.log(len(dlpred))
ohmapper = {
'A': [1, 0, 0, 0],
'C': [0, 1, 0, 0],
'G': [0, 0, 1, 0],
'T': [0, 0, 0, 1]
}
fivebase = seq[cutsite - 1]
onebp_features = ohmapper[fivebase] + [norm_entropy] + [del_score]
onebp_features = np.array(onebp_features).reshape(1, -1)
rate_1bpins = float(rate_model.predict(onebp_features))
# Predict 1 bp genotype frequencies
pred_1bpins_d = defaultdict(list)
for ins_base in bp_model[fivebase]:
freq = bp_model[fivebase][ins_base]
freq *= rate_1bpins / (1 - rate_1bpins)
pred_1bpins_d['Category'].append('ins')
pred_1bpins_d['Length'].append(1)
pred_1bpins_d['Inserted Bases'].append(ins_base)
pred_1bpins_d['Predicted_Frequency'].append(freq)
pred_1bpins_df = pd.DataFrame(pred_1bpins_d)
pred_df = pred_df.append(pred_1bpins_df, ignore_index=True)
pred_df['Predicted_Frequency'] /= sum(pred_df['Predicted_Frequency'])
join_cols = ['Category', 'Genotype Position', 'Length', 'Inserted Bases']
mdf = df.merge(pred_df, how='outer', on=join_cols)
mdf['Frequency'].fillna(value=0, inplace=True)
mdf['Predicted_Frequency'].fillna(value=0, inplace=True)
obs = mdf['Frequency']
pred = mdf['Predicted_Frequency']
r = pearsonr(obs, pred)[0]
alldf_dict['gt_r'].append(r)
obs_entropy = entropy(obs) / np.log(len(obs))
pred_entropy = entropy(pred) / np.log(len(pred))
alldf_dict['obs entropy'].append(obs_entropy)
alldf_dict['pred entropy'].append(pred_entropy)
alldf_dict['_Experiment'].append(exp)
alldf_dict['rs'].append(rs)
return alldf_dict
def calc_statistics(orig_df, exp, rate_model, bp_model, alldf_dict, rs):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
if sum(df['Count']) <= 1000:
return
df = orig_df
# Get observed frameshift rates
obs_fs = {'+0': 0, '+1': 0, '+2': 0}
all_ins_lens = set(df[df['Category'].isin(['ins',
'ins_notatcut'])]['Length'])
for ins_len in all_ins_lens:
crit = (df['Category'].isin(['ins', 'ins_notatcut'])) & (df['Length']
== ins_len)
fs = ins_len % 3
count = sum(df[crit]['Count'])
key = '+%s' % (int(fs))
obs_fs[key] += count
all_del_lens = set(df[df['Category'].isin(['del',
'del_notatcut'])]['Length'])
for del_len in all_del_lens:
crit = (df['Category'].isin(['del', 'del_notatcut'])) & (df['Length']
== del_len)
fs = (-1 * del_len) % 3
count = sum(df[crit]['Count'])
key = '+%s' % (int(fs))
obs_fs[key] += count
tot = sum(obs_fs.values())
for key in obs_fs:
obs_fs[key] /= tot
# Predict
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(orig_df)
# Predict rate of 1 bp insertions
# Featurize first
del_score = _predict2.total_deletion_score(seq, cutsite)
dlpred = _predict2.deletion_length_distribution(seq, cutsite)
norm_entropy = entropy(dlpred) / np.log(len(dlpred))
ohmapper = {
'A': [1, 0, 0, 0],
'C': [0, 1, 0, 0],
'G': [0, 0, 1, 0],
'T': [0, 0, 0, 1]
}
fivebase = seq[cutsite - 1]
onebp_features = ohmapper[fivebase] + [norm_entropy] + [del_score]
onebp_features = np.array(onebp_features).reshape(1, -1)
rate_1bpins = float(rate_model.predict(onebp_features))
# Predict 1 bp frequency
freq = rate_1bpins / (1 - rate_1bpins)
pred = list(dlpred)
pred.insert(0, freq)
pred = np.array(pred) / sum(pred)
pred_fs = {'+0': 0, '+1': 0, '+2': 0}
pred_fs['+1'] += pred[0]
for idx in range(1, len(pred)):
del_len = idx
fs = (-1 * del_len) % 3
key = '+%s' % (int(fs))
pred_fs[key] += pred[idx]
# Bae predict
bae_fs = {'+0': 0, '+1': 0, '+2': 0}
bae_dlpred = bae_prediction(seq, cutsite)
for idx in range(len(bae_dlpred)):
del_len = idx + 1
fs = (-1 * del_len) % 3
key = '+%s' % (int(fs))
bae_fs[key] += bae_dlpred[idx]
for fs in ['+0', '+1', '+2']:
alldf_dict['Frame'].append(fs)
alldf_dict['Bae'].append(bae_fs[fs])
alldf_dict['inDelphi'].append(pred_fs[fs])
alldf_dict['Obs'].append(obs_fs[fs])
alldf_dict['_Experiment'].append(exp)
alldf_dict['rs'].append(rs)
return alldf_dict
def calc_statistics(orig_df, exp, rate_model, bp_model, alldf_dict, rs):
# Calculate statistics on df, saving to alldf_dict
# Deletion positions
df = _lib.mh_del_subset(orig_df)
df = _lib.indels_without_mismatches_subset(df)
if sum(df['Count']) <= 1000:
return
df = orig_df
obs_dellens = []
# 1 bp insertion
crit = (df['Category'] == 'ins') & (df['Indel with Mismatches'] !=
'yes') & (df['Length'] == 1)
obs_dellens.append(sum(df[crit]['Count']))
# 1 - 28 bp deletions
for del_len in range(1, 28 + 1):
crit = (df['Category'] == 'del') & (df['Indel with Mismatches'] !=
'yes') & (df['Length'] == del_len)
obs_dellens.append(sum(df[crit]['Count']))
obs_dellens = np.array(obs_dellens) / sum(obs_dellens)
obs = obs_dellens
# Predict
_predict2.init_model()
seq, cutsite = _lib.get_sequence_cutsite(orig_df)
# Predict rate of 1 bp insertions
# Featurize first
del_score = _predict2.total_deletion_score(seq, cutsite)
dlpred = _predict2.deletion_length_distribution(seq, cutsite)
norm_entropy = entropy(dlpred) / np.log(len(dlpred))
ohmapper = {
'A': [1, 0, 0, 0],
'C': [0, 1, 0, 0],
'G': [0, 0, 1, 0],
'T': [0, 0, 0, 1]
}
fivebase = seq[cutsite - 1]
onebp_features = ohmapper[fivebase] + [norm_entropy] + [del_score]
onebp_features = np.array(onebp_features).reshape(1, -1)
rate_1bpins = float(rate_model.predict(onebp_features))
# Predict 1 bp frequency
freq = rate_1bpins / (1 - rate_1bpins)
pred = list(dlpred)
pred.insert(0, freq)
pred = np.array(pred) / sum(pred)
r = pearsonr(obs, pred)[0]
alldf_dict['dl_r'].append(r)
obs_entropy = entropy(obs) / np.log(len(obs))
pred_entropy = entropy(pred) / np.log(len(pred))
alldf_dict['obs entropy'].append(obs_entropy)
alldf_dict['pred entropy'].append(pred_entropy)
alldf_dict['_Experiment'].append(exp)
alldf_dict['rs'].append(rs)
return alldf_dict