Thoughts on a Smarter Planet is a special blogger series in partnership with leading IBM experts. Join the conversation as these experts discuss the innovations in science, business and systems like transportation that are helping build a Smarter Planet. About this program. 
Influenza is a moving target. The virus survives by accumulating mutations on its antigenic proteins that avoid recognition by neutralizing antibodies produced by the host immune system. It is these continual evolutions under intense immune selection that gives influenza the characteristics of rapid evolution, producing an ever-changing plethora of genetic variants that steadily render each vaccine ineffective.
Public health and pharmaceutical organizations engage in an annual exercise of developing seasonal flu vaccines. This current strategy, which I would characterize as being reactive, is comprised of the following steps:
Surveillance: Monitoring incidence of flu through health care provider organizations, and regional collection and analysis of samples.
Strain Characterization: The routine analysis of gathered samples to identify and characterize the particular genetic strains in circulation. Gene sequencing of novel strains often identifies new mutations.
Strain Selection: Public health agencies, such as FDA in the United States, study and select strain(s) that are prevalent in current season and likely to be the dominant circulating strains in next flu season.
Vaccine Production: Pharmaceutical companies under contract from public health agencies manufacture vaccines comprised of selected strains and market the vaccines to healthcare providers.
Effectiveness of the vaccines varies from year to year, as the viruses in the vaccine change each year based on international surveillance and scientists’ estimations about which types and strains of viruses will circulate in a given year.
For example, the composition of seasonal influenza vaccine for the Northern Hemisphere was announced this past February, composed of vaccines against three different virus strains. Certain years, for instance as recently as 2007/2008, the seasonal vaccine was a poor match for circulating strains and afforded little protection from infection.
The influenza virus exercises a few different mechanisms to rapidly create new genetic variations. Broadly speaking these are categorized into antigenic drift and antigenic shift. Drift is the accumulation of genetic mutations in the genes that code for viral proteins. Antigenic shift is the whole scale shuffling of large regions of the genome, between different strains of the virus.
Often, antigenic drift is responsible for viral variations from season to season. Therefore, the virus leaves a strong trail of data describing antigenic drift. Antigenic shift is harder to trace. On few occasions there is evidence of entirely new strains emerging, for example H5N1 (avian flu) in 2004, and H1N1 (swine flu) in 2009.
What does it take to be proactive? Can we ever successfully anticipate antigenic drift?
It is conceivable, in principle, to attempt doing so. Due to advances in technology outlined below, some recent and some imminent, the goal of predicting flu variations may be within reach in next 5 years. Some steps along the way are:
Genomic surveillance: With decreasing cost and increasing throughput of DNA sequencing technologies, it is conceivable to routinely practice in public health settings, whole genome sequencing of influenza virus from samples obtained via public health surveillance. This will allow rapid and continuous monitoring and complete mapping of genetic landscape being traversed by antigenic drift. In the future, advances in DNA sequencing would help in accelerating the pace, reducing costs and increasing the scope of genomic surveillance of infectious diseases.
Smart algorithms to predict antigenic drift: In recent work at IBM we analyzed gene sequences from 1968 to 2010 to model the evolutionary paths of influenza virus, which allows us to predict its potential antigenic drift. Results show that antigenic changes accumulate over time, with occasional large changes due to multiple co-occurring mutations at antigenic sites.
Smart models to anticipate escape from antibody neutralization, and switch in receptor specificity: We developed novel computational methodology performed on IBM Blue Gene supercomputers to study the effect of mutations. For example, such modeling shows, that a single mutation could render the vaccine ineffective. We also found that a double mutation could potentially allow the H5N1 avian flu virus to gain a foothold in human population.
Rapid screening of anticipated antigenic variants: Combining above prediction and modeling efforts with experimental means to rapidly screen predicted antigenic variants against libraries of antibodies could allow us to discover broadly neutralizing antibodies. More routinely, with adequate validation, this may inform the vaccine development process and allow the manufacture and stockpiling of vaccines against as yet unseen but potentially deadly variants of influenza.
Ajay Royyuru heads the Computational Biology Center at IBM Research, engaged in basic and exploratory research at the intersection of information technology and biology.
