Skip to main content

Table 2 Architecture of our proposed 2D CNN

From: Cancer type prediction based on copy number aberration and chromatin 3D structure with convolutional neural networks

Layer

Type

Output size

Conv (size, channel, pad)

Max pooling

input

in

176*176*ch

N/A

N/A

conv1

c + r + p

88*88*32

3*3, 32, 1

2*2

conv2

c + r + p

44*44*64

3*3, 64, 1

2*2

conv3

c + r + p

22*22*128

3*3128, 1

2*2

conv4

c + r + p

11*11*256

3*3, 256, 1

2*2

conv5

c + r

1*1*1024

11*11, 1024, 0

N/A

fc6

fc + r + d

1*1*1024

1*1, 1024, 0

N/A

fc7

fc

1*1*25

1*1, 25, 0

N/A

loss

sm + log

1*1

N/A

N/A

  1. Annotations - in: input layer; c: convolutional layer; r: ReLU layer; p: pooling layer; fc: fully connected layer; d: dropout layer; sm: softmax layer; log: log loss layer; ch: number of input channels (depending on whether the HiC data is used); asterisk(*): multiplication