DeepFlu: a deep learning approach for forecasting symptomatic influenza A infection based on pre-exposure gene...

Area of Science:

• Computational biology
• Genomics
• Infectious disease modeling

Background:

• Influenza A virus (IAV) infection outcomes vary among individuals.
• KLRD1 is a potential biomarker for flu susceptibility, but its predictive power is limited.
• Forecasting symptomatic flu infection based on pre-exposure gene expression remains an open question.

Purpose of the Study:

• To develop a predictive model for symptomatic flu infection using pre-exposure host gene expression.
• To assess the feasibility of forecasting influenza A virus (IAV) infection risk.
• To compare the performance of the developed model against existing biomarkers.

Main Methods:

• Developed DeepFlu, a deep learning model, utilizing human gene expression data from IAV-infected individuals (H1N1 and H3N2 subtypes).
• Employed deep neural networks, comparing their performance against convolutional neural networks, random forests, and support vector machines.
• Validated the model using leave-one-person-out cross-validation and external datasets.

Main Results:

• DeepFlu accurately forecasts symptomatic flu infection, indicating gene expression reflects host immunity.
• DeepFlu achieved high accuracy (70.0-73.8%) and AUC (0.700-0.847) in cross-validation and external validation, outperforming the KLRD1 biomarker.
• Pre-exposure data yielded the best predictive performance; combining H1N1 and H3N2 data did not significantly improve predictions.

Conclusions:

• DeepFlu serves as a prognostic tool for identifying individuals susceptible to influenza.
• The model demonstrates moderate capability in recognizing flu susceptibility.
• DeepFlu has the potential to aid in preventing the spread of influenza A virus (IAV).