Deep learning guided prediction modeling of dengue virus evolving serotype

Area of Science:

• Virology
• Genomics
• Computational Biology

Background:

• Viral evolution, particularly in Dengue Virus (DENV), drives disease emergence and impacts public health.
• Mutations in DENV serotypes can lead to genotypic alterations, increasing the risk of outbreaks.
• Accurate prediction of emerging DENV serotypes is vital for disease control and prevention.

Purpose of the Study:

• To develop and evaluate a deep learning model for forecasting emerging Dengue Virus serotypes.
• To leverage complete genome sequences for predicting DENV genomic alterations.
• To enhance the understanding of DENV evolution and its public health implications.

Main Methods:

• A deep learning model, DL-DVE, was developed using Long Short-Term Memory (LSTM) for generation and Feedforward Neural Network (FNN) for classification.
• The model was trained on 2000 publicly available DENV complete genome sequences.
• Conserved motifs were identified using MEME Suite, and sequence similarity was analyzed via BLAST.

Main Results:

• The DL-DVE model achieved 93% accuracy in sequence generation and high reliability in classification (ROC-AUC 0.818, ~99% for other metrics).
• Generated sequences were classified as DENV-4 with significant similarity to this serotype and distinctness from others.
• Analysis revealed conserved motifs and highlighted intra-serotype divergence, underscoring sequence uniqueness.

Conclusions:

• The DL-DVE model demonstrates proficiency in learning intricate genomic patterns for predicting emerging DENV serotypes.
• This deep learning approach offers a novel method for forecasting viral evolution and potential outbreaks.
• The findings contribute to a better understanding of DENV genomic diversity and can aid in developing targeted diagnostic and vaccine strategies.