Christina Wang and Maia Richards-Dinger, Department of Applied Mathematics, University of Colorado Boulder

Christina: mRNA sequence design via dynamic programming and tensor networks

Maia: Climate informed Bayesian forecasting of seasonal influenza dynamics using a non-stationary ARIMA model

Christina's Abstract: TBD

Maia's Abstract: 

Seasonal influenza recurs in temperate regions each winter, causing a major health burden. The timing and intensity of influenza outbreaks vary substantially across years, raising the need for accurate forecasts to guide public health messaging and inform resource allocation. Cold, dry weather enhances influenza transmission; thus, influenza forecasts may benefit from including temperature and humidity data. Some models have incorporated weather into their near-term predictions. However, this approach does not use climate variables to predict important seasonal parameters, such as peak week. We propose a novel forecasting approach that uses meteorological data to predict both the overall shape of an influenza season and its short-term dynamics. 

We developed a climate-driven non-stationary ARIMA model to predict influenza incidence at the US state level. We used ILI+, a combination of outpatient visits due to ILI and positive influenza test percentages, as a proxy for influenza incidence. For each year and state, we fit a piecewise linear function to the log of ILI+ and recorded the resulting values of our target parameters: up slope, down slope, peak week, and peak value. We fit a Bayesian multivariate linear model to predict these four parameters using climate and population variables. We compared models using all possible variable combinations. The best model included population density, percent of population under age 18, maximum temperature, and latitude. Higher temperatures correlated with later and lower peaks, and slower growth. To forecast future influenza seasons, we used predicted climate variables to create posterior distributions for our target parameters for each state. We forecast deviations from the predicted general trajectory using an ARIMA model.