ViroNia: LSTM based proteomics model for precise prediction of HCV

Area of Science:

• Virology
• Bioinformatics
• Machine Learning

Background:

• Accurate viral protein classification is crucial for understanding virus evolution and developing interventions.
• Existing classification methods face scalability limitations, hindering real-time analysis.

Purpose of the Study:

• To introduce ViroNia, a novel Long Short-Term Memory (LSTM)-based system for high-accuracy viral protein classification.
• To evaluate ViroNia's performance against other deep learning architectures and traditional methods like BLAST.

Main Methods:

• Development of ViroNia, an LSTM-based system utilizing pairwise sequence similarity and efficient data handling.
• Training and testing ViroNia on a dataset of 2250 viral protein sequences from NCBI and BVBRC databases.
• Performing fivefold cross-validation to assess classification accuracy.

Main Results:

• ViroNia achieved high accuracy rates of 99.7% (broad) and 99.6% (detail-level) classification.
• Cross-validation yielded average accuracies of 92.29% (±1.55%) for broad and 90.31% (±5.41%) for detail-level classification.
• ViroNia demonstrated superior performance compared to Simple RNN, GRU, 1D CNN, and Bidirectional LSTM models.

Conclusions:

• ViroNia offers a scalable, real-time solution for viral protein classification with automatic feature extraction.
• The system addresses limitations of traditional tools like BLAST, enabling efficient analysis of large viral datasets.
• ViroNia significantly contributes to viral research by enhancing classification accuracy and enabling rapid analysis.