Trans-Driver

Trans-Driver is a deep supervised learning framework utilizing a multilayer perceptron converter network with an unbalanced loss function. It integrates multi-omics data to learn both the differences and associations among diverse omics types for the identification of cancer driver genes. Compared with other state-of-the-art driver gene identification methods, Trans-Driver achieves excellent performance on benchmark datasets such as TCGA and CGC.Among ~20,000 protein-coding genes, Trans-Driver reported 269 candidate driver genes, of which 132 (about 49%) are present in the gold-standard CGC dataset. Further feature analysis demonstrated that integrating multi-omics data significantly improves predictive performance compared to using only somatic mutation data, and many identified candidates are clinically meaningful, confirming the practical utility of Trans-Driver.

Trans-Driver's dataset includes the model's training set, test set, and TCGA, CGC and PCAWG data, which you can find under ./data. Trans-Driver was evaluated using Fisher's test, and results for 33 cancers and comparison methods are in ./results. The Trans-Driver program includes model training and testing in ./program.

We can run the Trans-Driver model with the following command:

Usage:

python ./program/Trans-Driver.py [--lr LEARNING_RATE] [--batch_size BATCH_SIZE] [--epoch EPOCHS] [--alpha ALPHA] [--gamma GAMMA]

You can adjust the training process of Trans-Driver using several command-line arguments: --lr sets the learning rate, --batch_size specifies the number of samples per batch, --epoch controls the number of training epochs, --alpha defines the alpha parameter for the focal loss function (to balance class weights), and --gamma sets the gamma parameter for the focal loss (to focus learning on hard-to-classify samples).

Example:

python ./program/Trans-Driver.py --lr 0.00089 --batch_size 16 --epoch 30

You can then obtain the AUC and PR curves of Trans-Driver and the other comparison algorithms on the TCGA, CGC, and PCAWG datasets, as well as the enrichment analysis results comparing Trans-Driver with other methods on the TCGA and CGC datasets.

Enrichment Analysis on PCAWG:

python ./program/performance.py

Running this script evaluates the statistical enrichment of predicted driver genes by Trans-Driver on the PCAWG dataset, compared with a reference set of known driver genes.

Omics Contribution Analysis:

python ./program/contribution.py

The relative contributions of each omics modality were calculated and are available for both the TCGA pan-cancer dataset and each of the 33 individual cancer types.

Performance of MLP on TCGA, CGC, and PCAWG Datasets:

python ./program/mlp.py

The script generates both receiver operating characteristic (ROC) and precision-recall (PR) curves for the Multilayer Perceptron (MLP) model on the TCGA, CGC, and PCAWG datasets.

Trans-Driver's model implementation is based on Pytorch. its dependency packages are: Python (3.7.10), PyTorch (1.8.1), NumPy (1.19.5), Pandas (1.2.4), Keras (2.4.3), Scipy(1.6.2). The operating system is windows10. The CPU is an Intel Xeon Platinum 8255C (2.50 GHz), and the GPU is NVIDIA GeForce RTX 3090.