DNA-Inception

One of the hallmarks of cancer is somatic aberrations in the genomes and transcriptomes of malignant tumors, this information offers a wealth of distinguishable tumor-specific events, that do not occur in normal tissue of the same individual. Here, we propose DNA-Inception, a convolutional neural network-based model, it follows a sequence-based approach to classify AS events specific to tumor tissues. DNA-Inception takes as an input DNA or RNA sequences of AS events and their corresponding labels, i.e., tumor-specific 1 or tissue-specific 0. The following figure shows the architecture of the model:

       

Using this software

Installation

git clone https://github.com/nec-research/DNA-Inception.git

You can train the model using:

python ./src/train_model.py --help

An example input data format is provided in ./json

Contact

Israa Alqassem ([email protected])

Training and test datasets are available from:

• Kim, Pora, et al. "ExonSkipDB: functional annotation of exon skipping event in human." Nucleic acids research 48.D1 (2020): D896-D907.
• Kahles, André, et al. "Comprehensive analysis of alternative splicing across tumors from 8,705 patients." Cancer cell 34.2 (2018): 211-224.