mmostavi / cnncancertype Goto Github PK

I am working on a thesis, and want to download data from TCGA. I came across R scripts, but the downloaded data is a bit haywire. Is it possible for you to share the code of how you downloaded the data from TCGA?

@MMostavi , would it be possible to disclose the preprocessing code making the input data files?

@MMostavi, are the following files available for research purposes:

1. TCGA_new_pre_second.pckl
2. TCGA_new_pre_first.pckl

when i run 5cv_1D_CNN_33class.py， there is an error as follows.

Traceback (most recent call last): File "5cv_1D_CNN_33class.py", line 69, in <module> y_s = train_labels NameError: name 'train_labels' is not defined

Hello,

I was hoping to mimic your method for the 1D convolutional neural network cited in your paper. When I came to assess your code for the 1D CNN I was surprised to see that all the code was a 2D CNN. Instead of inputting a vector as stated in the paper, it seems like whats inputted is the same expected input as the 2D CNN. Therefore I am curious if the code provided is incorrect?
From your paper:
"This CNN model takes the gene expression as a vector and applies one-dimensional kernels to the input vector [....] The input for 1D-CNN (Fig. 1a) is a 1D vector following gene symbol’s alphabetic order, "

However per your code:

 X_cancer_samples_mat = np.reshape(X_cancer_samples_mat, (-1, 71, 100))
for train, test in kfold.split(X_cancer_samples_34, y_s):   # input_Xs in normal case and shuffled should be shuffled_Xs
input_Xs = input_Xs.reshape(input_Xs.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
input_Xs = input_Xs.astype('float32')

num_classes = len(onehot_encoded[0])

model = Sequential()
## *********** First layer Conv
model.add(Conv2D(32, kernel_size=(1, 71), strides=(1, 1),
                 input_shape=input_shape))

You reshape the vector per the wrapper method into a 71 x 100 2D input. Is there other code doing what you claim in the paper?

Edit:
It looks like the code for the 1D CNN is actually the code for the 2D Hybrid CNN