On February 28th, 2018 the National Institute of Allergy and Infectious Diseases (NIAID) released a strategic plan to develop a universal influenza vaccine. This plan comes after another severe flu season where the number of laboratory-confirmed influenza hospitalizations has now surpassed the last flu epidemic of 2014-15, reaching 67.9 hospitalizations per 100,000 people in the United States. Estimates released by the Centers for Disease Control and Prevention (CDC) indicate that this year’s vaccine is only 36% effective against influenza A and B. This estimate drops to 25% effective against the most common strain this season, H3N2.
To improve flu vaccine effectiveness in the future, the NIAID is seeking to develop a universal flu vaccine that is at least 75% effective, protects against influenza A viruses, confers at least one year of protection, and is suitable for all ages. To develop this universal vaccine, the NIAID will focus on the following critical areas of research:
- Clarifying how the influenza virus is transmitted.
- Identifying how protective influenza immunity occurs.
- Designing new ways to boost immunity.
Identifying biomarkers of protective influenza immunity
Given the history of the flu vaccine’s less-than-ideal effectiveness, novel approaches will be needed to fulfill these critical areas of research. Artificial intelligence (AI) and machine learning may be part of the solution. AI may soon be used to identify biomarkers of protective influenza immunity that will not only help us determine how protective immunity occurs, but also allow us to develop new ways to boost this immunity.
In 2017, Sanofi Pasteur and Berg partnered together to identify biomarkers of flu vaccine effectiveness. Sanofi Pasteur will provide Berg with data from its clinical trials on flu vaccines, while Berg will use its Interrogative Biology platform to analyze the data. This proprietary AI system can analyze biological information from patient samples along with clinical data to identify biomarkers of protective influenza immunity. Although this technology is still in its infancy, it has the potential to revolutionize the way we develop vaccines in the future.
Predicting the influenza season with artificial intelligence:
The CDC currently tracks flu activity via their influenza surveillance system. The goals of this system are to keep track of influenza-associated illnesses, hospitalizations, and deaths, as well as to determine the type of flu in circulation and any changes to the virus. However, the influenza surveillance system often lags behind real time flu trends, and is thus not entirely effective at informing public health decisions and preventing the flu.
The CDC acknowledged this gap back in 2013 and launched the “Predict the influenza season challenge.” The goal of this challenge was for participants to predict critical attributes of the 2013-14 flu season including timing, intensity, and the peak of the season. Results from the challenge’s first year highlighted that forecasting the flu was possible but that the accuracy was low and would require improvement. Ultimately, if we can accurately predict this information, we will be able to develop more effective countermeasures and reduce the number of influenza cases. In the years since the CDC started this challenge, machine learning has been used to try to forecast the flu and is still under development today.
Carnegie Mellon’s flu forecast models:
Carnegie Mellon’s flu forecasts have been the most accurate out of all entries for three years in a row. They utilize two models including their Delphi-stat and Delphi-epicast model to predict influenza trends.
The Delphi-stat model uses non-mechanistic statistical machine learning and data from previous influenza seasons to make predictions about future influenza seasons. The Delphi-epicast model uses a “wisdom of crowds” model to forecast future influenza seasons by incorporating predictions made by human participants. Interestingly, during the 2016-2017 challenge, Delphi-stat did better at predicting short-term trends while Delphi-epicast won out at forecasting long-term trends. These results highlight the value we can gain from AI in predicting influenza accurately in the future, but also show us that we should never underestimate our human intelligence.
Although the flu was extremely severe this season, we are living in an exciting time where researchers are developing systems based on AI and machine learning to better combat future outbreaks. These new techniques could allow us to create more efficient influenza vaccines and more accurately predict the timing and severity of the flu season.
