A Deep Learning Approach to the Screening of Oncogenic Gene Fusions in Humans

Area of Science:

• Computational biology
• Genomics
• Cancer research

Background:

• Gene fusions play a critical role in cancer development.
• Predicting the oncogenic potential of gene fusions is challenging.
• Existing methods rely on protein domain analysis, which is not easily adaptable to new data.

Purpose of the Study:

• To develop a flexible deep learning methodology for predicting the oncogenic probability of gene fusion transcripts.
• To categorize gene fusions as oncogenic or non-oncogenic.
• To overcome limitations of existing protein domain-based analysis.

Main Methods:

• Utilized raw protein sequences as input for a deep learning model.
• Employed Convolutional Neural Networks (CNNs) to infer oncogenicity probability scores.
• Trained and validated the model on a large dataset of pre-annotated gene fusions.

Main Results:

• The CNN-based method achieved an overall prediction accuracy of approximately 72% for gene fusion oncogenicity.
• Accuracy increased to 86% for high-confidence predictions.
• Demonstrated the flexibility of the deep learning approach for retraining with new data.

Conclusions:

• Deep learning, specifically CNNs, provides a powerful and adaptable tool for predicting gene fusion oncogenicity.
• The proposed method offers a more flexible and data-adaptable approach compared to traditional domain analysis.
• This methodology can be readily retrained for different cancer types, advancing cancer research.