TMO-Net is a pre-trained tumor multi-omics deep learning model to learn representation across multi-omics in cancers, improving the prediction performance of oncology tasks such as survival prediction or drug response.
TMO-Net was implemented by Pytorch and details are the followning:
- The source code of TMO-Net architecture can be found at the file of TMO_Net_model.py in 'model' module.
- The pipelines of pre-training and downstream task fine-tuning are available at the train/train_tcga_pancancer_multitask.py file.
- The implements of loss function are in the file of util/loss_function.py.
Overview of TMO-Net research, including pan-cancer multi-omics collection, TMO-Net model architecture, pre-training and fine-tuning, and biological interpretation analysis.
Multi-omics profiling of pre-training and downstream tasks were all downloaded from the public source. The processed data can be available at https://zenodo.org/records/10944664.
- Download the processed multi-omics profile from the url of zenodo and put them at the right directory.
- In the file of train/train_tcga_pancancer_multitask.py, we provide the training functions of TMO-Net pre-traing, pan-cancer classification and pan-cancer survival prediction with the default hyper-parameters. The examples setting of pre-training and fine-tuning TMO-Net are also displayed in the file. You can follow the examples to run TMO-Net.
This source code is licensed under the MIT license.