Skip to main content

Table 1 CORENup structure

From: CORENup: a combination of convolutional and recurrent deep neural networks for nucleosome positioning identification

Features extraction path

Layer

Kernel Dim

# Hidden Units

stride Dim

Output Dim

# Params

Conv1D

5

50

1

147x50

1.050

MaxPool1D

-

-

2

73x50

0

Dropout 50%

-

-

-

73x50

0

LSTM Path

Layer

Kernel Dim

# Hidden Units

stride Dim

Output Dim

# Params

LSTM

-

50

-

73x50

20.200

Dropout 50%

-

-

-

73x50

0

Flatten

-

-

-

3.650x1

0

Convolutional path

Layer

Kernel Dim

# Hidden Units

stride Dim

Output Dim

# Params

Conv1D

10

50

1

73x50

25.050

MaxPool1D

-

-

2

36x50

0

Dropout 50%

-

-

-

36x50

0

Flatten

-

-

-

1.800x1

0

Dense path

Concatenate

-

-

-

5.450x1

0

Dense

-

370

-

370x1

2.016.870

Dropout 50%

-

-

-

370x1

0

Dense

-

1

-

1x1

371

Parameters count

 

# Parameters

Features Extraction Path

1.050

LSTM Path

20.200

Convolutional Path

25.050

Dense Path

2.017.241

Total

2.063.541

  1. Notice that the dense layer contains the majority of the network parameters