Recently, the Bill & Melinda Gates Foundation announced it would award $12 million to accelerate the search for a universal flu vaccine. This is welcome news to researchers like me who are on the front lines of this effort. Any support to the larger cause moves us all that much closer to a solution.
In a Boston speech to the Massachusetts Medical Society, Bill Gates, Microsoft founder and philanthropist, painted the stakes clearly and starkly. He said that the work to find a universal flu vaccine is part of the global effort to more effectively stem the risks of a “deadly global pandemic” of any kind.
In making this announcement, Mr. Gates said he is working in conjunction with Google co-founder Larry Page and his wife Lucy.
To be sure, if the 2017–18 flu season taught us anything, it’s that our conventional approach to vaccination can be easily overcome by a new or unpredictable virus strain.
In the ensuing public dialogue on this announcement, one question emerged that points to the real potential for the development of a universal flu vaccine, which is, “How might we use artificial intelligence and big data to help scientists advance our understanding?” That understanding, presumably, would be used to accelerate the creation of the universal flu vaccine.
A significant part of the answer is through the deployment of an operations research (OR) approach, combined with analytics, to better capture and process the vast amount of data that already exist and are continually generated. As patterns are quickly analyzed and better understood, we can narrow our focus on the right solutions that benefit specific segments of the population, while leading toward larger protections for everyone.
In recent years, researchers have partnered with leading universities and organizations across the country to make significant progress toward solutions. In 2015, a team of multidisciplinary investigators involving immunology, genomics and advanced analytics from the Emory Vaccine Center, the U.S. Centers for Disease Control and Prevention (CDC) and the Georgia Institute of Technology, of which I am grateful and honored to have been a part, were awarded the INFORMS Daniel H. Wagner Prize for Excellence in Operations Research Practice for their groundbreaking work in predicting the immunity of a vaccine without exposing individuals to infection.
This approach addresses a longstanding challenge in the development of vaccines—that of only being able to determine immunity or effectiveness long after vaccination and often only after being exposed to infection. The team has identified predictive gene signatures for several deadly infectious diseases, including yellow fever, malaria and influenza. These predictive signatures could guide the rapid development of vaccines against emerging infections and aid in the monitoring of suboptimal immune responses in the elderly, infants or people with weakened immune systems.
As the lead analyst of this remarkable team, and in order to tackle the complex and vast biological and genomic data, I employed OR and analytics methods to design a general-purpose machine learning framework, DAMIP. This framework enabled us to capture massive amounts of data from a number of sources to facilitate the rapid design and evaluation of new and emerging vaccines. We were also able to identify individuals most likely to not be protected by a vaccine. DAMIP identifies gene signatures that can predict vaccine immunity and efficacy. It is unique in its ability to identify and utilize the most important data sets, and to cancel out unnecessary “noise,” while bringing clarity to previously gray areas in vaccine development.
With the DAMIP model, the results for yellow fever demonstrated for the first time that a vaccine’s ability to immunize a patient could be predicted successfully (with accuracy of greater than 90 percent) within one week after vaccination. DAMIP also identifies genes for predicting whether someone will produce high levels of antibodies against a flu shot a few days after vaccination. After scanning the extent to which carefully selected genes are turned on in white blood cells, the researchers can predict on day three, with up to 90 percent accuracy, who will make high levels of antibodies against a standard flu shot four weeks after vaccination.
It often takes several weeks after vaccination for an individual to develop sufficient levels of protective antibodies against the influenza virus. The ability to predict who will meet these criteria within a few days after vaccination and identify non-responders is of great value from a public health perspective. These predictive signatures could guide the rapid development of vaccines against emerging infections. In particular, the common genes identified across multiple influenza strands could be important in the design of a universal flu vaccine.
Pandemic flu is a serious global health concern. Advances in influenza virology, immunology and vaccinology are critical to the development of a universal influenza vaccine. This has proven particularly challenging, as there are multiple approaches that include triggering antibodies that can react with multiple flu subtypes, or stimulating T cells, which are actively involved in immune system response, though this solution is complicated by a lack of knowledge on what biological marker is best for immune protection. Many current vaccine designs also use HAI titer (a specific amount of flu antibodies), though numerous studies have shown that HAI titer is a poor correlate for vaccine immune protection.
Operations research and analytics can play an important role in assisting experimental scientists in their study of viruses or design of vaccines. For example, in our work, we identify the best immune correlate(s) for protection, which facilitates a better design for flu vaccine, including universal ones. When studying the malaria virus, we identified markers for predicting clinical outcome, those who can be protected by the vaccine or those high-risk individuals who would not benefit from it.
Such computational approaches can also assist in understanding the mutation of the viruses, evolutionary trends, genetic distances and diversity, and geographical distribution. Understanding how some viruses mutate while others sustain through long periods of time will help us better comprehend the gene flow and may help assess and predict future strands and the risks that they carry. By utilizing analytics, the results can generate biological hypotheses and may influence the direction of experiments.
Of course, for now, people should continue to get seasonal flu shots when they are available. But be assured that there is a broad and concerted ongoing effort to develop the means to most efficiently target the right and most vulnerable people in society for the right vaccines, while pursuing our larger goal of a universal flu vaccine.