def plot_seq_importance(grads,
x,
xlim=None,
ylim=None,
layer_idx=-2,
figsize=(25, 3),
title=""):
"""Plot sequence importance score
Args:
grads: either deeplift or gradientxinput score matrix
x: one-hot encoded DNA sequence
xlim: restrict the plotted xrange
figsize: matplotlib figure size
"""
grads = grads.squeeze()
x = x.squeeze()
seq_len = x.shape[0]
vals_to_plot = grads * x
if xlim is None:
xlim = (0, seq_len)
if ylim is None:
ylim = (np.amin(vals_to_plot), np.amax(vals_to_plot))
seqlogo_fig(vals_to_plot, figsize=figsize)
plt.xticks(list(range(xlim[0], xlim[1], 5)))
plt.xlim(xlim)
plt.ylim(ylim)
plt.title(title)
def manual_test_layer_plots_AA():
motifs = ["ACDEFGGIKNY"]
seq = encodeAA(motifs)
seq_length = 100
motif_width = 7
seqlogo_fig(seq[0], vocab="AA")
plt.show()
# specify the input shape
input_dna = cl.InputAA(seq_length)
# convolutional layer with filters initialized on a PWM
x = ConvAA(
filters=1,
kernel_size=motif_width, # motif width
activation="relu",
# mean_max_scale of 1 means that only consensus sequence gets score larger than 0
)(input_dna)
# Smoothing layer - positional-dependent effect
x = cl.GAMSmooth(n_bases=10, l2_smooth=1e-3, l2=0)(x)
x = cl.GlobalSumPooling1D()(x)
x = kl.Dense(units=1, activation="linear")(x)
model = Model(inputs=input_dna, outputs=x)
model.compile("adam", "mse")
# TODO - test
model.layers[1].plot_weights(plot_type="heatmap")
model.layers[1].plot_weights(0, plot_type="motif_raw")
model.layers[1].plot_weights(0, plot_type="motif_pwm_info")
def _plot_weights_motif(self, index, plot_type="motif_raw",
background_probs=DEFAULT_BASE_BACKGROUND,
ncol=1,
figsize=None):
"""Index can only be a single int
"""
w_all = self.get_weights()
if len(w_all) == 0:
raise Exception("Layer needs to be initialized first")
W = w_all[0]
if index is None:
index = np.arange(W.shape[2])
if isinstance(index, int):
index = [index]
fig = plt.figure(figsize=figsize)
if plot_type == "motif_pwm" and plot_type in self.AVAILABLE_PLOTS:
arr = pssm_array2pwm_array(W, background_probs)
elif plot_type == "motif_raw" and plot_type in self.AVAILABLE_PLOTS:
arr = W
elif plot_type == "motif_pwm_info" and plot_type in self.AVAILABLE_PLOTS:
quasi_pwm = pssm_array2pwm_array(W, background_probs)
arr = _pwm2pwm_info(quasi_pwm)
else:
raise ValueError("plot_type needs to be from {0}".format(self.AVAILABLE_PLOTS))
fig = seqlogo_fig(arr, vocab=self.VOCAB_name, figsize=figsize, ncol=ncol, plot_name="filter: ")
# fig.show()
return fig
def plot_seq_importance(model, x, xlim=None, layer_idx=-2, figsize=(25, 3)):
"""Plot input x gradient sequence importance score
Args:
model: DNA-sequence based Sequential keras model
x: one-hot encoded DNA sequence
xlim: restrict the plotted xrange
figsize: matplotlib figure size
"""
seq_len = x.shape[1]
if xlim is None:
xlim = (0, seq_len)
grads = input_grad(model, x, layer_idx=layer_idx)
for i in range(len(x)):
seqlogo_fig(grads[i]*x[i], figsize=figsize)
plt.xticks(list(range(xlim[0], xlim[1], 5)))
plt.xlim(xlim)
def plotPWMInfo(self, figsize=(10, 2)):
pwm = self.pwm
info = _pwm2pwm_info(pwm)
fig = seqlogo_fig(info, vocab="DNA", figsize=figsize)
plt.ylabel("Bits")
return fig
def plot_seq_importance(scores,
data,
ylim=None,
figsize=(25, 3),
outf=None,
tick_interval=5):
seq_len = data.shape[0]
product = np.expand_dims(scores, axis=1) * data
product = np.nan_to_num(product, copy=False)
print(product.shape)
seqlogo_fig(product, figsize=figsize)
plt.xticks(list(range(0, product.shape[0], tick_interval)))
if ylim != None:
plt.ylim(ylim)
if outf == None:
plt.show()
else:
plt.axis("off")
plt.gca().set_position([0, 0, 1, 1])
plt.savefig(outf)
def plot_seq_importance(grads, x, xlim=None, ylim=None, layer_idx=-2, figsize=(25, 3),title="",snp_pos=0):
"""Plot sequence importance score
Args:
grads: either deeplift or gradientxinput score matrix
x: one-hot encoded DNA sequence
xlim: restrict the plotted xrange
figsize: matplotlib figure size
"""
grads=grads.squeeze()
x=x.squeeze()
seq_len = x.shape[0]
vals_to_plot=grads*x
if xlim is None:
xlim = (0, seq_len)
if ylim is None:
ylim= (np.amin(vals_to_plot),np.amax(vals_to_plot))
seqlogo_fig(vals_to_plot, figsize=figsize)
plt.xticks(list(range(xlim[0], xlim[1], 5)))
plt.xlim(xlim)
plt.ylim(ylim)
plt.title(title)
plt.axvline(x=snp_pos, color='k', linestyle='--')
def plotPSSM(self, background_probs=DEFAULT_BASE_BACKGROUND, figsize=(10, 2)):
pssm = self.get_pssm()
return seqlogo_fig(pssm, vocab="DNA", figsize=figsize)
def plotPWM(self, figsize=(10, 2)):
pwm = self.pwm
fig = seqlogo_fig(pwm, vocab="DNA", figsize=figsize)
plt.ylabel("Probability")
return fig