Predicting the Antigenic Evolution of Influenza Viruses

(This is post 3 of 5 on the ‘Limits to prediction’ ACtioN (applied complexity network) meeting organised by the Santa Fe Institute (SFI) and hosted by Willis Towers Watson on 9 September 2016.) While some talks explored the limits of what can be predicted Dr Terry Jones presented a case study on how designing vaccines for the influenza virus was practical only once its future adaptations could be predicted. In essence it is a case study of applying imagination (see post on Fundamental limits to prediction) in order to transform an impossible task into a simple task. The talk began with some history of the impact of influenza. Notably while most people’s day-to-day experience with the influenza virus (the flu) is relatively benign this has not always been the case. Between 1918-1920 a flu pandemic spread across the world infecting 500 million people (33% of the world’s population, and including remote islands in the Pacific and the Arctic) and killing somewhere between 50 to 100 million of those infected (3-6% of the population). To put this in perspective, World War I resulted in the death of 17 million people. Given the possibility of global pandemics, to have such a virus circulating in the population means there is a great deal of interest in studying it. As the flu is a virus, vaccination is required to stop it spreading through a population. This is where problems are encountered. It takes time (9-12 months) to produce a vaccine in sufficient amounts to vaccinate those most vulnerable to the flu and the virus is continually evolving (making past vaccines ineffective). The challenge is therefore to produce a vaccine that is effective on the flu strain that will be circulating when the vaccine is being administered – but which doesn’t exist yet. Dr Jones showed how his team have taken the data used by those trying to make this prediction (with varying degrees of success) and transformed it into an “antigenic map” (essentially the proteins produced by the virus). Mutations can occur in any protein, making the search impossibly large. The breakthrough was the realisation that a virus is harmless if it cannot bind to a human cell, and therefore only the mutations in a tiny subset of the proteins matter. The map showed the groupings of strains with similar antigenic properties corresponding to groups of flu strains that shared common vaccines and how these antigenic properties would be stable for a short time before evolving into (jumping to) a new state – making past vaccines ineffective. This mapping also made prediction of where the virus might “jump” next possible and therefore it became possible to develop a vaccine with sufficient lead time that was likely to be successful in protecting people against the flu. From an investment point of view the talk showed how things that might seem unpredictable may be predicted, or, at least become less unpredictable, if the right properties are looked at with the right perspective. While financial risk models are typically very good at analysing our current portfolio (if tomorrow looks like yesterday) they are poor at incorporating how a portfolio (and by extension the investment strategy that creates it) will change in response to future market conditions. If we wish to better protect (immunise) our portfolio against negative events maybe more time should be spent thinking about how our strategy and economies/markets evolve (the long term changes invisible to our risk models) and less about what these look like today. And finally, it reminded us that pandemics and other “extreme risks” do occur from time to time. If investing assets on behalf of future generations with a multi-generational investment horizon then some low probability, high impact events are a matter of when and not if.