Estimating the time-varying effective reproduction number via Cycle Threshold-based Transformer

Area of Science:

• Epidemiology
• Infectious Disease Dynamics
• Computational Biology

Background:

• Effective monitoring of infectious disease spread is crucial for timely public health interventions.
• Traditional incidence-based methods for estimating the effective reproduction number (Rt) have limitations, including data biases and reliance on early-stage epidemic information.
• Viral load data, specifically cycle threshold (Ct) values, offer a potential alternative for inferring epidemic trajectories.

Purpose of the Study:

• To develop and evaluate a novel deep learning model, the Cycle Threshold-based Transformer (Ct-Transformer), for estimating the time-varying effective reproduction number (Rt).
• To assess the performance of the Ct-Transformer compared to traditional incidence-based and existing Ct-based Rt estimation methods.
• To investigate the robustness of the Ct-Transformer to varying detection resource levels and explore its utility in self-supervised learning scenarios.

Main Methods:

• Development of the Cycle Threshold-based Transformer (Ct-Transformer) model utilizing cycle threshold (Ct) values from infected populations.
• Supervised learning approach to train the Ct-Transformer for Rt estimation.
• Application of self-supervised pre-training followed by fine-tuning for Rt estimation.
• Validation using both synthetic and real-world infectious disease datasets.

Main Results:

• The supervised Ct-Transformer significantly outperformed traditional incidence-based statistical methods and existing Ct-based Rt estimation techniques.
• The Ct-Transformer demonstrated robustness to variations in detection resources, a key advantage over traditional methods.
• Self-supervised pre-training followed by fine-tuning achieved performance comparable to the supervised Ct-Transformer.
• The Ct-based deep learning approach improved real-time Rt estimates, showing adaptability to new and emerging epidemics.

Conclusions:

• The Ct-Transformer provides a more accurate and robust method for real-time estimation of infectious disease transmission (Rt) compared to conventional approaches.
• Deep learning models leveraging viral load data (Ct values) offer a promising avenue for enhanced epidemic monitoring and response.
• The Ct-Transformer's adaptability and robustness make it a valuable tool for tracking newly emerging infectious diseases.