Skip to content

Latest commit

 

History

History
108 lines (87 loc) · 6.46 KB

README.md

File metadata and controls

108 lines (87 loc) · 6.46 KB
Raw

MOViDA

Introduction

Our tool makes drug sensitivity prediction for cancer cell lines using a XAI model based on gene ontology hierarchy. We use multi-omics data, including mutations, amplifications, deletions and pathway activity to make predictions. This cutting-edge approach allows for a more accurate prediction of drug sensitivity by taking into account the specific genetic makeup of the cancer cell line. By utilizing gene ontology hierarchy and multi-omics data, our tool provides a powerful resource for researchers and clinicians looking to improve treatment outcomes for patients with cancer.

Environment set up

  • Hardware: GPU with CUDA >= 11
  • Software:
    • Python >= 3.7
    • Miniconda
    • PyTorch >= 11.3 (verify on Pytorch the corrected version for your pc) 
      conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
    • Python libraries
      • networkx
      • numpy
      • pandas
      • torchmetrics
      • torchvision

Files

  • code
    • main.py : starts training or testing
    • param_data.py : contains a class that represents the data in input
    • param_train.py : contains a class that represents the training parameters
    • prepare_directories.py : creates directory and file for the experiment
    • models : directory that contains classes which encode all the parameters and structure of the models, as well as functions needed for training them
      • MOViDA.py : class for drug sensitivity prediction
      • MOViDA_synergy.py : class for synergistic drug combination prediction
    • parameters : directory that contains .txt files for data and parameters information, see below

Parameters

A parameters file is a .txt file that contains a Python dictionary with all the necessary information for data and training. An example of such a file would be:

{
    "input_dir"         : "data", 
    "names_mo"          : ["mut", "amp", "del"], 
    ...
}

  • data parameters : This is a python dictionary that contains the input data needed to launch a training process. The dictionary contains the following key-value pairs :

    • "input_dir": This is the path to the directory where the input data files are located.
    • "names_mo": This is a list of strings that contains the names of the different types of genetic modifications (e.g. "mut", "amp", "del")
    • "drug2id": This is the path to the file that contains the mapping from drug names to drug IDs.
    • "cell2id": This is the path to the file that contains the mapping from cell line names to cell line IDs.
    • "mo_gene2id": This is a list of strings that contains the paths to the files that map the gene names to the genetic modification IDs for each type of modification.
    • "train_file": This is the path to the file that contains the training set of drug-cell line pairs.
    • "val_file": This is the path to the file that contains the validation set of drug-cell line pairs.
    • "test_file": This is the path to the file that contains the test set of drug-cell line pairs.
    • "all_file": This is the path to the file that contains all the drug-cell line pairs.
    • "cell2features": This is a list of strings that contains the paths to the files that contain the features of the cell lines for each type of modification.
    • "names_drfeat": This is a list of strings that contains the names of the different types of drug features (e.g. "pc", "vs").
    • "drug2features": This is a list of strings that contains the paths to the files that contain the features of the drugs for each type of feature.
    • "pathway_act_file": This is the path to the file that contains the pathway activity data.
    • "ontology": This is the path to the file that contains the ontology data.
    • "mo_gene2ontology": This is a list of strings that contains the paths to the files that map the genes to the ontology for each type of modification.
    • "synergy_bool": This is a boolean value that indicates whether the training process should include synergy information (i.e. whether the output should predict synergy scores or not).

  • training parameters : This is a python dictionary that contains the training parameters needed to launch a training process :

    • "epoch": This is the number of training iterations.
    • "lr": This is the learning rate used for training.
    • "batchsize": This is the number of samples in a batch used for training.
    • "loss_func": This is the name of the loss function used for training.
    • "metric_func": This is the name of the metric used to evaluate the performance of the model.
    • "classifier": This is a boolean value that indicates whether the training process is for classification or regression.
    • "wloss_bool": This is a boolean value that indicates whether the loss function is weighted or not.
    • "focal_bool": This is a boolean value that indicates whether the Focal Loss is used or not.
    • "f_alpha": This is the value used for Focal Loss alpha.
    • "f_gamma": This is the value used for Focal Loss gamma.
    • "sampler_bool": This is a boolean value that indicates whether the data should be oversampled or not.
    • "ccl_hiddens": This is the number of nodes in hidden layers in the cell line encoder.
    • "drug_hiddens": This is a string that contains the number of hidden layers in the drug encoder and how many nodes they have.
    • "final_hiddens": This is the number of nodes in hidden layers in the final network.
    • "max_val": This is the maximum value used for the target variable.
    • "eps_weights": This is the parameter epsilon using to balance the weights.
    • "expdir": This is the directory where all the experiments results will be stored.
    • "currexpdir": This is the directory where the current experiment results will be stored, it will be in "expdir".
    • "result_file": This is the path to the file where the test set predictions will be stored.
    • "result_all_file": This is the path to the file where the all set predictions will be stored.
    • "model_name": This is the name of python file storing the model used (without the file extension .py)
    • "load_model": This is the path to the pre-trained model if any.

Start training

  • prepare dictionary files for data and parameter
  • execute this command whit the correct data_dict and param_dict 
    python code/main.py -data_dict data_dict -param_dict param_dict -cuda 0