DeepM6A: Detection and Eclucidation of DNA Methylation on N6-Adenine
DeepM6A application implements a deep-learning-based algorithm for predicting potential DNA 6mA sites de novo from sequence at single-nucleotide resolution. Application of this tool to three representative model organisms, Arabidopsis thaliana, Drosophila melanogaster and Escherichia coli, demonstrated that it supercedes the conventional k-mer-based approaches in both accuracy and performance.
References:
- Fei Tan, Tian Tian, Xiurui Hou, Xiang Yu, Lei Gu, Fernanda Mafra, Brian D. Gregory, Zhi Wei and Hakon Hakonarson,
Elucidation of DNA methylation on N6-adenine with deep learning
Nature Machine Intelligence (2020), Volume 2, Pages 466–475. https://doi.org/10.1038/s42256-020-0211-4.
Documentation
Important Notes
- Module Name: DeepM6A (see the modules page for more information)
- Unusual environment variables set
- DM6A_HOME installation directory
- DM6A_BIN executable directory
- DM6A_DIR source code directory
- DM6A_DATA sample data directory
- DM6A_MODELS pre-trained models directory
Interactive job
Interactive jobs should be used for debugging, graphics, or applications that cannot be run as batch jobs.
Allocate an interactive session and run the program. Sample session:
[user@biowulf]$ sinteractive --mem=16g --gres=gpu:p100:1,lscratch:10 -c4 [user@cn3104 ~]$ module load DeepM6A/0.1 [+] Loading CUDA Toolkit 9.0.176 ... [+] Loading cuDNN 7.0 libraries... [+] Loading DeepM6A 0.1 [user@cn3104 ~]$ cp -r $DM6A_DATA Data [user@cn3104 ~]$ cp -r $DM6A_MODELS ModelsThe training, validation and testing data in the Data folder come as HDF5 files. To learn more about the format of these data, execute the commands:
[user@cn3104 ~]$ python
>>> import h5py
>>> trainmat = h5py.File('Data/m6A_30.train_1.hdf5', 'r')
>>> import numpy as np
>>> X_train = np.transpose(np.array(trainmat['x_train']),axes=(0,2, 1))
>>> X_train.shape
(31406, 61, 4)
>>> y_train = np.array(trainmat['y_train'])
>>> y_train.shape
(31406,)
Here is how the data can be used for training a model:
>>> from DEM6A import DeepM6A
>>> DA = DeepM6A('./', epochs=5)
>>> DA.fit(train_name='Data/m6A_30.train_1.hdf5', valid_name='Data/m6A_30.valid_1.hdf5')
train_label: count (array([0, 1]), array([15703, 15703]))
valid_label: count (array([0, 1]), array([1962, 1962]))
building model...............
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/DEM6A/deepm6a.py:68: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(activation="linear", name="conv1", input_shape=(61, 4), filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal")`
name = "conv1"))
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/DEM6A/deepm6a.py:79: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(activation="linear", name="conv2", filters=80, kernel_size=2, strides=1, padding="valid", kernel_initializer="he_normal")`
name = "conv2"))
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/DEM6A/deepm6a.py:90: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(activation="linear", name="conv3", filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal")`
name="conv3"))
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/DEM6A/deepm6a.py:101: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(activation="linear", name="conv4", filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal")`
name = "conv4"))
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/DEM6A/deepm6a.py:113: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(padding="valid", activation="linear", name="conv5", filters=80, kernel_size=4, strides=1, kernel_initializer="he_normal")`
name = "conv5"))
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv1 (Conv1D) (None, 58, 80) 1360
_________________________________________________________________
leaky_re_lu_1 (LeakyReLU) (None, 58, 80) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 58, 80) 0
_________________________________________________________________
conv2 (Conv1D) (None, 57, 80) 12880
_________________________________________________________________
leaky_re_lu_2 (LeakyReLU) (None, 57, 80) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 57, 80) 0
_________________________________________________________________
conv3 (Conv1D) (None, 54, 80) 25680
_________________________________________________________________
leaky_re_lu_3 (LeakyReLU) (None, 54, 80) 0
_________________________________________________________________
dropout_3 (Dropout) (None, 54, 80) 0
_________________________________________________________________
conv4 (Conv1D) (None, 51, 80) 25680
_________________________________________________________________
leaky_re_lu_4 (LeakyReLU) (None, 51, 80) 0
_________________________________________________________________
dropout_4 (Dropout) (None, 51, 80) 0
_________________________________________________________________
conv5 (Conv1D) (None, 48, 80) 25680
_________________________________________________________________
leaky_re_lu_5 (LeakyReLU) (None, 48, 80) 0
_________________________________________________________________
dropout_5 (Dropout) (None, 48, 80) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 3840) 0
_________________________________________________________________
dense_1 (Dense) (None, 100) 384100
_________________________________________________________________
leaky_re_lu_6 (LeakyReLU) (None, 100) 0
_________________________________________________________________
dropout_6 (Dropout) (None, 100) 0
_________________________________________________________________
dense_2 (Dense) (None, 1) 101
_________________________________________________________________
activation_1 (Activation) (None, 1) 0
=================================================================
Total params: 475,481
Trainable params: 475,481
Non-trainable params: 0
_________________________________________________________________
compiling and fitting model...........
Train on 31406 samples, validate on 3924 samples
Epoch 1/5
2020-08-18 16:33:51.512423: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2020-08-18 16:33:51.695354: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1432] Found device 0 with properties:
name: Tesla P100-PCIE-16GB major: 6 minor: 0 memoryClockRate(GHz): 1.3285
pciBusID: 0000:91:00.0
totalMemory: 15.90GiB freeMemory: 15.64GiB
2020-08-18 16:33:51.695432: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
2020-08-18 16:33:52.111563: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2020-08-18 16:33:52.111639: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988] 0
2020-08-18 16:33:52.111655: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0: N
2020-08-18 16:33:52.111800: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 15155 MB memory) -> physical GPU (device: 0, name: Tesla P100-PCIE-16GB, pci bus id: 0000:91:00.0, compute capability: 6.0)
- 4s - loss: 0.6410 - acc: 0.6241 - val_loss: 0.4876 - val_acc: 0.7757
Epoch 00001: val_loss improved from inf to 0.48762, saving model to ./DeepM6A.hdf5
Epoch 2/5
- 1s - loss: 0.4618 - acc: 0.7918 - val_loss: 0.3504 - val_acc: 0.8626
Epoch 00002: val_loss improved from 0.48762 to 0.35043, saving model to ./DeepM6A.hdf5
Epoch 3/5
- 1s - loss: 0.4064 - acc: 0.8249 - val_loss: 0.3194 - val_acc: 0.8726
Epoch 00003: val_loss improved from 0.35043 to 0.31939, saving model to ./DeepM6A.hdf5
Epoch 4/5
- 1s - loss: 0.3818 - acc: 0.8396 - val_loss: 0.3052 - val_acc: 0.8807
Epoch 00004: val_loss improved from 0.31939 to 0.30522, saving model to ./DeepM6A.hdf5
Epoch 5/5
- 1s - loss: 0.3675 - acc: 0.8476 - val_loss: 0.3004 - val_acc: 0.8907
Epoch 00005: val_loss improved from 0.30522 to 0.30042, saving model to ./DeepM6A.hdf5
training done!
To perform prediction, use a pre-trained model from the Models folder. For example:
>>> DA.predict('Data/m6A_30.test_1.hdf5', 'results.csv', 'Models/Ecoli/bestmodel.hdf5')
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(batch_input_shape=[None, 61,..., name="cov1", activity_regularizer=None, trainable=True, input_dtype="float32", activation="linear", input_shape=(61, 4), filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_1", rate=0.2)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(name="cov2", activity_regularizer=None, trainable=True, activation="linear", input_shape=(None, Non..., filters=80, kernel_size=2, strides=1, padding="valid", kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_2", rate=0.2)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(name="cov3", activity_regularizer=None, trainable=True, activation="linear", input_shape=(None, Non..., filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_3", rate=0.2)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(name="cov4", activity_regularizer=None, trainable=True, activation="linear", input_shape=(None, Non..., filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_4", rate=0.2)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Conv1D` call to the Keras 2 API: `Conv1D(name="cov5", activity_regularizer=None, trainable=True, activation="linear", input_shape=(None, Non..., filters=80, kernel_size=4, strides=1, padding="valid", kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_5", rate=0.5)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(name="dense_1", activity_regularizer=None, trainable=True, input_dim=None, activation="linear", units=100, kernel_initializer="he_normal", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dropout` call to the Keras 2 API: `Dropout(trainable=True, name="dropout_6", rate=0.5)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/base_layer.py:1109: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(name="dense_2", activity_regularizer=None, trainable=True, input_dim=None, activation="linear", units=1, kernel_initializer="glorot_uniform", kernel_regularizer=None, bias_regularizer=None, kernel_constraint=None, bias_constraint=None, use_bias=True)`
return cls(**config)
/usr/local/apps/DeepM6A/0.1/lib/python3.6/site-packages/keras/engine/saving.py:327: UserWarning: Error in loading the saved optimizer state. As a result, your model is starting with a freshly initialized optimizer.
warnings.warn('Error in loading the saved optimizer '
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
cov1 (Conv1D) (None, 58, 80) 1360
_________________________________________________________________
leakyrelu_1 (LeakyReLU) (None, 58, 80) 0
_________________________________________________________________
dropout_1 (Dropout) (None, 58, 80) 0
_________________________________________________________________
cov2 (Conv1D) (None, 57, 80) 12880
_________________________________________________________________
leakyrelu_2 (LeakyReLU) (None, 57, 80) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 57, 80) 0
_________________________________________________________________
cov3 (Conv1D) (None, 54, 80) 25680
_________________________________________________________________
leakyrelu_3 (LeakyReLU) (None, 54, 80) 0
_________________________________________________________________
dropout_3 (Dropout) (None, 54, 80) 0
_________________________________________________________________
cov4 (Conv1D) (None, 51, 80) 25680
_________________________________________________________________
leakyrelu_4 (LeakyReLU) (None, 51, 80) 0
_________________________________________________________________
dropout_4 (Dropout) (None, 51, 80) 0
_________________________________________________________________
cov5 (Conv1D) (None, 48, 80) 25680
_________________________________________________________________
leakyrelu_5 (LeakyReLU) (None, 48, 80) 0
_________________________________________________________________
dropout_5 (Dropout) (None, 48, 80) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 3840) 0
_________________________________________________________________
dense_1 (Dense) (None, 100) 384100
_________________________________________________________________
leakyrelu_6 (LeakyReLU) (None, 100) 0
_________________________________________________________________
dropout_6 (Dropout) (None, 100) 0
_________________________________________________________________
dense_2 (Dense) (None, 1) 101
_________________________________________________________________
activation_1 (Activation) (None, 1) 0
=================================================================
Total params: 475,481
Trainable params: 475,481
Non-trainable params: 0
_________________________________________________________________
test_label: count (array([0, 1]), array([1967, 1967]))
**************prediction results on test dataset************
3934/3934 [==============================] - 0s 93us/step
[3.8706200020307278, 0.4974580579562786]
3934/3934 [==============================] - 0s 76us/step
3934/3934 [==============================] - 0s 40us/step
************************
auc: 0.4892284721287104
mcc: -0.0222572017594063
precision:0.40384615384615385 recall:0.010676156583629894 f1score:0.020802377414561663 support:1967
************************
End the interactive session:
>>> quit() [user@cn3104 ~]$ exit salloc.exe: Relinquishing job allocation 46116226 [user@biowulf ~]$