def test_all_layers(seq, encodeSEQ, InputSEQ, ConvSEQ, tmpdir):
seq_length = len(seq[0])
# pre-process
train_x = encodeSEQ(seq)
train_y = np.array([[1]])
print(train_x.shape)
# build model
inp = InputSEQ(seq_length=seq_length)
if ConvSEQ == cl.ConvSplines:
x = ConvSEQ(filters=1)(inp)
else:
x = ConvSEQ(filters=1, kernel_size=1)(inp)
x = cl.GlobalSumPooling1D()(x)
m = Model(inp, x)
m.summary()
m.compile("adam", loss="mse")
m.fit(train_x, train_y)
filepath = str(tmpdir.mkdir('data').join('test_keras.h5'))
print(tmpdir)
m.save(filepath)
m = load_model(filepath)
assert isinstance(m, Model)
def manual_test_layer_plots_RNA():
motifs = ["TTAATGA"]
pwm_list = [PWM.from_consensus(motif) for motif in motifs]
seq_length = 100
motif_width = 7
# specify the input shape
input_dna = cl.InputDNA(seq_length)
# convolutional layer with filters initialized on a PWM
x = ConvRNA(
filters=1,
kernel_size=motif_width, # motif width
activation="relu",
kernel_initializer=ci.PSSMKernelInitializer(pwm_list),
bias_initializer=ci.PSSMBiasInitializer(pwm_list,
kernel_size=motif_width,
mean_max_scale=1)
# 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 test_convDNA(tmpdir):
motifs = ["TTAATGA"]
pwm_list = [PWM.from_consensus(motif) for motif in motifs]
seq_length = 100
motif_width = 7
# specify the input shape
input_dna = cl.InputDNA(seq_length)
# convolutional layer with filters initialized on a PWM
x = cl.ConvDNA(
filters=1,
kernel_size=motif_width, # motif width
activation="relu",
kernel_initializer=ci.PWMKernelInitializer(pwm_list),
bias_initializer=ci.PWMBiasInitializer(pwm_list,
kernel_size=motif_width,
mean_max_scale=1)
# 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)
# compile the model
model.compile(optimizer="adam", loss="mse", metrics=[cm.var_explained])
# filepath = "/tmp/model.h5"
filepath = str(tmpdir.mkdir('data').join('test_keras.h5'))
model.save(filepath)
m = load_model(filepath)
assert isinstance(m, Model)
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 get_pool(pooling):
pooling = deepcopy(pooling)
pool_type = pooling["pool_type"]
del pooling["pool_type"]
if pool_type == "max":
return [kl.MaxPooling1D(pool_size=pooling["pool_size"]), kl.Flatten()]
elif pool_type == "mean":
return [
kl.AveragePooling1D(pool_size=pooling["pool_size"]),
kl.Flatten()
]
elif pool_type == "weight":
return [cl.SmoothPositionWeight(**pooling), cl.GlobalSumPooling1D()]
else:
raise ValueError("")
def manual_test_layer_plots_Codon():
motifs = ["TTAATGAAT"]
seq_length = 102
# specify the input shape
input_dna = cl.InputCodon(seq_length)
# convolutional layer with filters initialized on a PWM
x = cl.ConvCodon(
filters=1,
kernel_size=2, # 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")
model.layers[1].plot_weights(figsize=(3, 10))
def test_ConvSplines(tmpdir):
x_pos = np.vstack([np.arange(15), np.arange(15)])
y = np.arange(2)
x = encodeSplines(x_pos)
inl = cl.InputSplines(15, 10)
o = cl.ConvSplines(
1,
kernel_regularizer=cr.GAMRegularizer(l2_smooth=.5),
)(inl)
o = cl.GlobalSumPooling1D()(o)
model = Model(inl, o)
model.compile("Adam", "mse")
model.fit(x, y)
filepath = str(tmpdir.mkdir('data').join('test_keras.h5'))
# load and save the model
model.save(filepath)
m = load_model(filepath)
assert isinstance(m, Model)
def single_layer_pos_effect(
pooling_layer="sum", # 'sum', 'max' or 'mean'
nonlinearity="relu", # 'relu' or 'exp'
motif_length=9,
n_motifs=6, # number of filters
step_size=0.01,
num_tasks=1, # multi-task learning - 'trans'
n_covariates=0,
seq_length=100, # pre-defined sequence length
# splines
n_splines=None,
share_splines=False, # should the positional bias be shared across motifs
# regularization
lamb=1e-5, # overall motif coefficient regularization
motif_lamb=1e-5,
spline_lamb=1e-5,
spline_param_lamb=1e-5,
# initialization
init_motifs=None, # motifs to intialize
init_motif_bias=0,
init_sd_motif=1e-2,
init_sd_w=1e-3, # initial weight scale of feature w or motif w
**kwargs): # unused params
# initialize conv kernels to known motif pwm's
if init_motifs:
# WARNING - initialization is not the same as for Concise class
pwm_list = [PWM.from_consensus(motif) for motif in init_motifs]
kernel_initializer = ci.PWMKernelInitializer(pwm_list,
stddev=init_sd_motif)
bias_initializer = ci.PWMBiasInitializer(pwm_list,
kernel_size=motif_length)
else:
# kernel_initializer = "glorot_uniform"
kernel_initializer = ki.RandomNormal(stddev=init_sd_motif)
bias_initializer = ki.Constant(value=init_motif_bias)
activation = nonlinearity # supports 'relu' out-of-the-box
# define the model
# ----------------
inputs = []
seq_input = kl.Input((seq_length, 4))
inputs.append(seq_input)
# convolution
xseq = kl.Conv1D(
filters=n_motifs,
kernel_size=motif_length,
kernel_regularizer=kr.l1(l=motif_lamb), # Regularization
activation=activation,
kernel_initializer=kernel_initializer,
bias_initializer=bias_initializer)(seq_input)
# optional positional effect
if n_splines:
xseq = cl.GAMSmooth(
n_bases=n_splines,
share_splines=share_splines,
l2_smooth=spline_lamb,
l2=spline_param_lamb,
)(xseq)
# pooling layer
if pooling_layer is "max":
xseq = kl.pooling.GlobalMaxPooling1D()(xseq)
elif pooling_layer is "mean":
xseq = kl.pooling.GlobalAveragePooling1D()(xseq)
elif pooling_layer is "sum":
xseq = cl.GlobalSumPooling1D()(xseq)
else:
raise ValueError("pooling_layer can only be 'sum', 'mean' or 'max'.")
# -----
# add covariates
if n_covariates:
cov_input = kl.Input((n_covariates, ))
inputs.append(cov_input)
x = kl.concatenate([xseq, cov_input])
else:
x = xseq
# -----
predictions = kl.Dense(
units=num_tasks,
kernel_regularizer=kr.l1(lamb),
kernel_initializer=ki.RandomNormal(stddev=init_sd_w))(x)
model = Model(inputs=inputs, outputs=predictions)
model.compile(optimizer=Adam(lr=step_size), loss="mse", metrics=["mse"])
return model
def model(
train_data,
activation="relu",
kernel_size=10,
filters=16,
conv2_use_skip=False,
internal_pos={"name": "global_maxpool"},
# {"name": "strided_maxpool", "pool_size": 3}
# {"name": "maxpool+rnn_sequence", "pool_size": 3, "dropout": 0.1}
# {"name": "rnn", "pool_size": 3, "dropout": 0.1}
# {"name": "maxpool+weight_sum", "pool_size": 3, "n_bases": 10, "share_splines": False,
# "l2_smooth": 1e-5, "l2": 0}
external_pos=None, # None, {"type": "gam", "as_track": True, "units": 1}
dropout_rate=0.5,
n_hidden=100,
use_batchnorm=False,
use_weightnorm=False,
lr=0.001):
"""Returns keras model for modelling arrays returned by data.data()
"""
# config
seq_length = train_data[0]["seq"].shape[1]
n_tasks = train_data[1].shape[1]
ext_n_bases = train_data[0]["dist_tss"].shape[2]
activation = PReLU() if activation == "PReLU" else activation
inputs = []
# position module
# ---------------
if external_pos is not None:
# conf
external_pos["as_track"] = train_data[0]["dist_tss"].shape[1] != 1
if external_pos["as_track"]:
pos_length = seq_length
else:
pos_length = 1
ext_filters = external_pos["units"]
# NOTE: Concise now implements a layer SplineT, which simplifies
# the following code significantly.
pos_inputs, pos_outputs = tuple(
zip(*[
pos_module(pos_length=pos_length,
ext_n_bases=ext_n_bases,
ext_filters=ext_filters,
kernel_size=kernel_size,
feat_name=feat_name,
ext_pos_kwargs=external_pos) for feat_name in
external_pos.get("feat_names", ["gene_start", "gene_end"])
]))
inputs += list(pos_inputs)
pos_outputs = list(pos_outputs)
# sequence module
# ----------------
# initialize conv kernels to known motif pwm's
seq_input = cl.InputDNA(seq_length, name="seq")
inputs += [seq_input]
x_pwm = cl.ConvDNA(filters=filters,
kernel_size=kernel_size,
activation=activation)(seq_input)
if use_batchnorm:
x_pwm = kl.BatchNormalization(axis=1)(x_pwm)
# inject external_pos as a track
if external_pos is not None and external_pos["as_track"]:
x_pwm = kl.concatenate([x_pwm] + pos_outputs, axis=-1)
x = kl.Conv1D(filters, kernel_size=1, activation=activation)(x_pwm)
if conv2_use_skip:
x = kl.concatenate([x_pwm, x]) # skip connection ?
if use_batchnorm:
x = kl.BatchNormalization(axis=1)(x)
# summarize across sequence
# -------------------------
if internal_pos["name"] == "global_maxpool":
x = kl.GlobalMaxPool1D()(x)
elif internal_pos["name"] == "strided_maxpool":
x = kl.MaxPool1D(pool_size=internal_pos["pool_size"])(x)
x = kl.Flatten()(x)
elif internal_pos["name"] == "maxpool+rnn_sequence":
x = kl.MaxPool1D(pool_size=internal_pos["pool_size"])(x)
x = kl.Bidirectional(
kl.LSTM(filters,
dropout_W=internal_pos["dropout"],
dropout_U=internal_pos["dropout"],
return_sequences=True))(x)
x = kl.Flatten()(x)
elif internal_pos["name"] == "rnn":
x = kl.MaxPool1D(pool_size=internal_pos["pool_size"])(x)
x = kl.Bidirectional(
kl.LSTM(filters,
dropout_W=internal_pos["dropout"],
dropout_U=internal_pos["dropout"]))(x)
elif internal_pos["name"] == "maxpool+weight_sum":
x = kl.MaxPool1D(pool_size=internal_pos["pool_size"])(x)
x = cl.SmoothPositionWeight(n_bases=internal_pos.get("n_bases", 10),
share_splines=internal_pos.get(
"share_splines", False),
l2_smooth=internal_pos.get("l2_smooth", 0),
l2=internal_pos.get("l2", 0))(x)
x = cl.GlobalSumPooling1D()(x)
else:
raise ValueError("invalid internal_pos")
if use_batchnorm:
x = kl.BatchNormalization()(x)
x = kl.Dropout(dropout_rate)(x)
# append external_pos as a scalar
# -------------------------------
if external_pos is not None and not external_pos["as_track"]:
x = kl.concatenate([x] + pos_outputs, axis=-1)
# FC layers
# ---------
x = kl.Dense(n_hidden, activation=activation)(x)
if use_batchnorm:
x = kl.BatchNormalization()(x)
x = kl.Dropout(dropout_rate)(x)
outputs = kl.Dense(n_tasks, activation="sigmoid")(x)
# compile model
# -------------
m = Model(inputs, outputs)
if use_weightnorm:
optimizer = AdamWithWeightnorm(lr=lr)
else:
optimizer = Adam(lr=lr)
m.compile(optimizer=optimizer, loss="binary_crossentropy", metrics=["acc"])
if use_weightnorm:
data_based_init(m, model_data.subset(train_data, np.arange(500))[0])
return m
