Tracking and forecasting influenza virus evolution

Trevor Bedford (@trvrb)
29 Aug 2019
Options X

Tracking seasonal influenza virus evolution

Population turnover of A/H3N2 influenza is extremely rapid


Project to provide a real-time view of the evolving influenza population

All in collaboration with Richard Neher

Nextflu Nextstrain

Real-time tracking of pathogen evolution

with Richard Neher, James Hadfield, Emma Hodcroft, Thomas Sibley, Colin Megill, John Huddleston, Barney Potter, Sidney Bell, Louise Moncla, Charlton Callender, Misja Ilcisin, Kairsten Fay, Jover Lee

Current view of H3N2 from

Forecasting has been a goal for a while

At Options IX in 2016, I stated:

We predict the 171K clade will continue to be successful (unless supplanted by a novel mutant)

Instead, momentum behind 171K (clade 3c2.A1) waned and co-circulation has increased

What's going on with this co-circulation?

Looking historically shows this to be rare

Variation in both diversity and rate of HI drift

Possible correlation of increased diversity when drift is slow

Risk of eventual "speciation"

Forecasting strain turnover

with John Huddleston and Richard Neher

Fitness models project strain frequencies

Future frequency $x_i(t+\Delta t)$ of strain $i$ derives from strain fitness $f_i$ and present day frequency $x_i(t)$, such that

$$\hat{x}_i(t+\Delta t) = x_i(t) \, \mathrm{exp}(f_i \, \Delta t)$$

Total strain frequencies at each timepoint are normalized. This captures clonal interference between competing lineages.

Two inputs

  • Estimate of present-day strain frequencies $x(t)$
  • Estimate of present-day strain fitnesses $f$

Strain frequency estimated via region-weighted KDE

Strain fitness estimated from viral attributes

The fitness $f$ of strain $i$ is estimated as

$$\hat{f}_i = \beta^\mathrm{A} \, f_i^\mathrm{A} + \beta^\mathrm{B} \, f_i^\mathrm{B} + \ldots$$

where $f^A$, $f^B$, etc... are different standardized viral attributes and $\beta^A$, $\beta^B$, etc... coefficients are trained based on historical evolution

Antigenic drift Intrinsic fitness Recent growth
epitope mutations non-epitope mutations local branching index
HI titers DMS data (via Bloom lab) delta frequency

Future population depends on frequency and fitness

Forecast assessed based on weighted distance match to observed future population

Forecast assessed based on weighted distance match to observed future population

Train in 6-year sliding windows from 1995 to 2015 with most recent years held out as test

Within-category performance favors HI drift, non-epitope fitness and local branching index

Composite model relies on local branching index and non-epitope fitness

Model successfully predicts clade growth

Best pick from model is generally close to best possible retrospective pick

Forecast from current virus population

Since March these two clades have begun to spread throughout the world

Predicted sequence match of circulating strains to future population

Goal is to start rolling these forecasts out live to shortly

Future work

  • Better predictors for antigenic drift
  • Incorporate NA evolution
  • Incorporate geography
  • Forecast H1N1pdm, Vic and Yam


Bedford Lab: Alli Black, John Huddleston, Barney Potter, James Hadfield, Katie Kistler, Louise Moncla, Maya Lewinsohn, Thomas Sibley, Jover Lee, Kairsten Fay, Misja Ilcisin

This work: WHO Global Influenza Surveillance Network, GISAID, John Huddleston, Richard Neher, Barney Potter, James Hadfield, Dave Wentworth, Becky Garten, Marta Łuksza, Michael Lässig, Richard Reeve, Jackie Katz, Colin Russell, John McCauley, Rod Daniels, Kanta Subbarao, Ian Barr, Aeron Hurt, Tomoko Kuwahara, Takato Odagiri



  • Lab website:
  • Flu tracking:
  • Twitter: @trvrb
  • Slides: