Skip to main content

Table 1 Detailed configurations of the method

From: Deep mixed model for marginal epistasis detection and population stratification correction in genome-wide association studies

Collector (1D-CNN)

Convolutional layer

Num. of Kernels: 16

Kernel Size: 1000 x 1

Padding: Same

  

Initializer: Truncated normal initializer

Activation: ReLU

 

Pooling layer

Size: 2000

Stride: 2000

 
 

1st fully-connected layer

Output: 32

Dropout rate: 0.9

 
 

2nd Fully-connected layer

Output: 1

  

Selector (LSTM)

Weighting layer

Num. of units: p (one-to-one layer)

 

Hidden layer

Num. of units: 0.15p

Optimizer (ADAM)

Learning rate: 0.001

Batch size: 128

  

Other hyperparams

Collector’s epoch: 20

Selector’s epoch: 1500

  
  1. The architecture and hyperparameters are selected through the experiments with simulated data, and are used without changes for real data experiments