This is to run neoantigen intrinsic feature-based deep learning model (neoDL) for identifying IDH wild type GBMs with better prognosis who will most likely benefit from neoantigen based personalized immunetherapy. neoDL is built with three hidden layers including two LSTM layers and one fully connected layers, each layer containing 128, 32 and 8 nodes, respectively. sigmoid function was chosen as neuron activation function for fully connected layers, MSE as the loss function and Adam as the iterative optimizer. The maximum number of iterations was set as 1000. The initial connection weights and biases of each layer were randomly generated, and end up reaching stable parameters through training iterations.

Click here to download neoDL code.
You can also visit https://github.com/zhangjbig/neoDL to use neoDL.

Python packages “keras”, “pandas” and “lifelines” are in need. Before installing “keras” package, make sure other relevant packages “tensorflow”, “numpy”, “scipy” and “matplotlib” are already installed. If not, run the following commands to install the above python packages respectively. The version of “tensorflow” we used is “1.13.0-rcl”.

`pip install tensorflow`
`pip install numpy`
`pip install scipy`
`pip install matplotlib`

Run the following command to install “keras”. The version of “keras” we used is “2.2.4”.

`pip install keras`


Run the following command to install “pandas”. The version of “pandas” we used is “0.25.3”.

`pip install pandas`


Run the following command to install “lifelines”. The version of “lifelines” we used is “0.24.3”.

`pip install lifelines`


LOOCV is adopted to train neoDL. Specifically, the training data was separated into two sections randomly with proportion of training and testing sets as 6 to 4. The training set was used to train the model to determine the unknown parameters, while the test set was used to validate the effect of the predicted parameters. To obtain the optimal model, the above process was carried out 300 times. Kaplan-Meier survival analysis was operated each time to see if the model can divide the samples into two groups with a statistically significant survival difference. Only groups with P-value lower than or equal to the threshold of 0.05 were regarded as statistically significant. Among 300 times trial, the more significant stratifications, the more stable our model is.

Train neoDL:
`python train_model.py -o <path> -i <input>`



Parameters:

-o: output directory
-i: input file (neoantigen intrinsic feature)


Creat a working directory, e.g., ./working
Put the DL trained model file (LSTM_model.h5 is the model we trained for IDH wild type GBM) and neoDL.py in the working directory
Run neoDL:
`python neoDL.py -o <path> -i <input> -m <model>`


Parameters:

-o: output directory
-i: input file (neoantigen intrinsic feature)
-m: trained DL model (for IDH wild type GBM, LSTM_model.h5)



Input file (neoantigen intrinsic feature) is a csv file, in which each row is a sample, and columns consist of sample names(denoted as sample), living days(denoted as days), vital status(denoted as vital, 1 represents for living and 0 represents for dead), mutation numbers(denoted as Num_mutations), and neoantigen intrinsic features.

The output includes cluster results, p-values from survival analysis and survival analysis plots. Cluster results will be written into the input csv file as an extra column, two groups are denoted as 0 and 1 respectively. Those files will be saved into a folder named “testnresult”. Survival analysis plots will be output as png files, and saved in a file named “surv_fig”. P-value will be directly output in the command window.

Citation:

Sun, T., He, Y., Li, W., et al. neoDL: a novel neoantigen intrinsic feature-based deep learning model identifies IDH wild-type glioblastomas with the longest survival. BMC Bioinformatics, 2021.22(1): 382. doi:10.1186/s12859-021-04301-6