def plot_motif(motif_name, figsize, ylab='bits', information_content=True):
"""
Plot motifs from encode motifs file
"""
motif_letter_heights = loaded_motifs.getPwm(motif_name).getRows()
return plot_pwm(motif_letter_heights, figsize,
ylab=ylab, information_content=information_content)
def get_motif_scores(encoded_sequences,
motif_names,
max_scores=None,
return_positions=False,
GC_fraction=0.4):
"""
Computes pwm log odds.
Parameters
----------
encoded_sequences : 4darray
motif_names : list of strings
max_scores : int, optional
return_positions : boolean, optional
GC_fraction : float, optional
Returns
-------
(num_samples, num_motifs, seq_length) complete score array by default.
If max_scores, (num_samples, num_motifs*max_scores) max score array.
If max_scores and return_positions, (num_samples, 2*num_motifs*max_scores)
array with max scores and their positions.
"""
encoded_sequences = np.transpose(encoded_sequences, (0, 1, 3, 2))
num_samples, _, _, seq_length = encoded_sequences.shape
scores = np.ones((num_samples, len(motif_names), seq_length))
for j, motif_name in enumerate(motif_names):
pwm = loaded_motifs.getPwm(motif_name).getRows().T
log_pwm = np.log(pwm)
gc_pwm = 0.5 * np.array(
[[1 - GC_fraction, GC_fraction, GC_fraction, 1 - GC_fraction]] *
len(pwm[0])).T
gc_log_pwm = np.log(gc_pwm)
scores[:, j, :] = get_pssm_scores(encoded_sequences,
log_pwm) - get_pssm_scores(
encoded_sequences, gc_log_pwm)
if max_scores is not None:
sorted_scores = np.sort(scores)[:, :, ::-1][:, :, :max_scores]
if return_positions:
sorted_positions = scores.argsort()[:, :, ::-1][:, :, :max_scores]
return np.concatenate(
(sorted_scores.reshape(
(num_samples, len(motif_names) * max_scores)),
sorted_positions.reshape(
(num_samples, len(motif_names) * max_scores))),
axis=1)
else:
return sorted_scores.reshape(
(num_samples, len(motif_names) * max_scores))
else:
return scores
def get_motif_scores(encoded_sequences,
motif_names,
max_scores=None,
return_positions=False,
GC_fraction=0.4):
"""
Computes pwm log odds.
Parameters
----------
encoded_sequences : 4darray
motif_names : list of strings
max_scores : int, optional
return_positions : boolean, optional
GC_fraction : float, optional
Returns
-------
(num_samples, num_motifs, seq_length) complete score array by default.
If max_scores, (num_samples, num_motifs*max_scores) max score array.
If max_scores and return_positions, (num_samples, 2*num_motifs*max_scores)
array with max scores and their positions.
"""
num_samples, _, _, seq_length = encoded_sequences.shape
scores = np.ones((num_samples, len(motif_names), seq_length))
for j, motif_name in enumerate(motif_names):
pwm = loaded_motifs.getPwm(motif_name).getRows().T
log_pwm = np.log(pwm)
gc_pwm = 0.5 * np.array(
[[1 - GC_fraction, GC_fraction, GC_fraction, 1 - GC_fraction]] * len(
pwm[0])).T
gc_log_pwm = np.log(gc_pwm)
scores[:, j, :] = get_pssm_scores(encoded_sequences,
log_pwm) - get_pssm_scores(
encoded_sequences, gc_log_pwm)
if max_scores is not None:
sorted_scores = np.sort(scores)[:, :, ::-1][:, :, :max_scores]
if return_positions:
sorted_positions = scores.argsort()[:, :, ::-1][:, :, :max_scores]
return np.concatenate(
(sorted_scores.reshape((num_samples, len(motif_names) * max_scores)),
sorted_positions.reshape(
(num_samples, len(motif_names) * max_scores))),
axis=1)
else:
return sorted_scores.reshape((num_samples, len(motif_names) * max_scores))
else:
return scores
def get_motif_scores(encoded_sequences,
motif_names,
max_scores=None,
return_positions=False,
GC_fraction=0.4,
pfm=None,
log_pfm=None,
include_rc=True):
"""
Computes pfm log odds.
Parameters
----------
encoded_sequences : 4darray
motif_names : list of strings
max_scores : int, optional
return_positions : boolean, optional
GC_fraction : float, optional
pfm: position weight matrix for the motif, optional
log_pfm: log(pfm), optional, this is the format that HOCOMOCO Provides in their PFM download links
include_rc: boolean indicating whether both the forward strand and the reverse complement of the motif should be used (default True)
Returns
-------
(num_samples, num_motifs, seq_length) complete score array by default.
If max_scores, (num_samples, num_motifs*max_scores) max score array.
If max_scores and return_positions, (num_samples, 2*num_motifs*max_scores)
array with max scores and their positions.
"""
encoded_sequences = np.transpose(encoded_sequences, (0, 1, 3, 2))
num_samples, _, _, seq_length = encoded_sequences.shape
scores = np.ones((num_samples, len(motif_names), seq_length))
for j, motif_name in enumerate(motif_names):
if (pfm is None) and (log_pfm is None):
pfm = loaded_motifs.getPwm(motif_name).getRows().T
log_pfm = np.log(pfm)
elif log_pfm is None:
log_pfm = np.log(pfm)
#get the background pfm either based on GC fraction or on shuffling the input sequence
background_pfm = 0.5 * np.array(
[[1 - GC_fraction, GC_fraction, GC_fraction, 1 - GC_fraction]] *
len(log_pfm[0])).T
background_log_pfm = np.log(background_pfm)
scores[:, j, :] = get_pssm_scores(
encoded_sequences, log_pfm,
include_rc=include_rc) - get_pssm_scores(
encoded_sequences, background_log_pfm, include_rc=include_rc)
if max_scores is not None:
sorted_scores = np.sort(scores)[:, :, ::-1][:, :, :max_scores]
if return_positions:
sorted_positions = scores.argsort()[:, :, ::-1][:, :, :max_scores]
return np.concatenate(
(sorted_scores.reshape(
(num_samples, len(motif_names) * max_scores)),
sorted_positions.reshape(
(num_samples, len(motif_names) * max_scores))),
axis=1)
else:
return sorted_scores.reshape(
(num_samples, len(motif_names) * max_scores))
else:
return scores