RF-models
| CNN_Model | Size |
| IT+LC CNN (1) | 1.36 GB |
| IT+LC CNN (2) | 1.36 GB |
| IT+LC CNN (3) | 1.36 GB |
Structure of the CNN
The structure of the CNN is demonstrated below. We leveraged the tensor flow embedded function of summary(). The numbers of parameters and the shape are shown below:
________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 90, 87, 64) 320
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 90, 43, 64) 0
_________________________________________________________________
conv2d_1 (Conv2D) (None, 89, 42, 128) 32896
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 89, 21, 128) 0
_________________________________________________________________
flatten (Flatten) (None, 239232) 0
_________________________________________________________________
dense (Dense) (None, 512) 122487296
_________________________________________________________________
dense_1 (Dense) (None, 2) 1026
=================================================================
Total params: 122,521,538
Trainable params: 122,521,538
Non-trainable params: 0
_________________________________________________________________
Experimental Parameters settings
Platform: GTX-2080ti Intel(R) Core(TM) i7-9700 CPU 64GB DDR4
CNN parameters:
epoch=50, batch_size=2, categorical_crossentropy,
optimizer=RMSprop,Learning rate=0.000002,metrics=accuracy
RandomForestClassifier: (n_estimators=120)
Kernal PCA:
KernelPCA (n_components=Feature_dimension, kernel=’poly’, gamma=0.15, degree=2)
Feature_dimension(lncRNA/ RF): 64-4096 (4096 selected)
___________________________________________________________________
Parameters concerning Repetitions of CV
| Evaluation inside dataset | Negative sampling RS (File Random seeds FRS)repetition | Dataset Divide Random seeds (DRS)Repetition | Trials |
| Type1 | 5 | 5 | 25 |
| Type2 (5folds Cross-validation) | 5 | 5 | 25 |
| Type3 (10-fold Cross-validation) | 5 | 5 | 25 |
| Type3 (HC involve) | 3 | 5 | 15 |
| Negative sampling Rate (NR Exploration) | 3 | 3 | 9 |
Parameters concerning Repetitions of Transfer Verification
| Prediction Parts | Train File Repetition | Test File Repetition | Metrics Calculation |
| Transfer Verification (HC, not involved) | 5 | 5 | 20 （25-5） |
| Transfer Verification (HC, involved) | 3 | 3 | 6 (9-3) |