Prediction of Proto-Oncogene Using Bidirectional GRU and Attention.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Proteomics
• Machine Learning in Biology

Background:

• Advancements in genome sequencing necessitate accurate protein sequence prediction.
• Proto-oncogene (OG) mutations and tumor suppressor gene (TSG) dysregulation drive uncontrolled cancer growth.
• Computational methods can identify OG/TSG-linked genes for targeted cancer drug development.

Purpose of the Study:

• To develop and validate deep learning models for predicting proto-oncogene protein sequences.
• To leverage attention mechanisms for improved protein sequence classification.
• To identify potential therapeutic targets for cancer treatment.

Main Methods:

• Proposed two deep learning models: Attention with Convolutional Neural Network (ACNN) and Attention with Bi-directional Gated Recurrent Units (ABiGRU).
• Utilized attention mechanisms for enhanced protein sequence classification.
• Validated models using independent testing, K-fold cross-validation, ablation studies, and statistical significance testing.

Main Results:

• The ACNN model achieved 96.85% accuracy on the benchmark Uniprot dataset.
• The ABiGRU model demonstrated higher accuracy at 97.53% on the benchmark Uniprot dataset.
• Both models showed superior performance through rigorous validation methods.

Conclusions:

• The developed attention-based deep learning models are effective for proto-oncogene protein sequence prediction.
• These models can aid in early cancer patient prognosis.
• The findings support the identification of novel cancer-fighting strategies and therapeutic targets.