CBIL-VHPLI: a model for predicting viral-host protein-lncRNA interactions based on machine learning and transfer...

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Virus-host protein-lncRNA interactions (VHPLI) are crucial for understanding viral pathogenesis and host immune responses.
• Previous VHPLI prediction studies have focused on plants and animals, with limited research on viral interactions.

Purpose of the Study:

• To develop a novel deep learning model for predicting VHPLI, specifically focusing on viral interactions.
• To enhance the accuracy and applicability of VHPLI prediction through transfer learning.

Main Methods:

• A deep learning model, CBIL-VHPLI, integrating convolutional neural networks (CNN) and bidirectional long and short-term memory (BiLSTM) networks with transfer learning.
• Feature extraction using k-mer, one-hot encoding, CTD, and Z curve methods for protein and lncRNA sequences.
• Model pretraining on diverse datasets followed by fine-tuning on viral-human lncRNA interactions.

Main Results:

• The pre-trained CBIL-VHPLI model achieved an accuracy of approximately 0.9 on external validation datasets.
• Fine-tuning on a viral protein-human lncRNA dataset resulted in an improved accuracy of 0.946.
• The model demonstrated a 91.6% prediction reproducibility rate with RIP-Seq experimental data and successfully predicted PIK3CD-AS2 and H5N1 NS1 interactions.

Conclusions:

• CBIL-VHPLI represents a significant advancement in predicting viral-host protein-lncRNA interactions.
• The model's high accuracy and experimental validation highlight its potential for elucidating molecular mechanisms in virology.
• The developed model and datasets are publicly available for academic research.