Illinois professor develops mathematical models to predict viral spreading patterns, develop better vaccines

Get a flu shot, prevent the flu? While the vaccine often works, there are other years when a high percentage of patients who received the shot get influenza anyway.

Prof. Olgica Milenkovic has developed new mathematical models to predict viral spreading patterns and help develop better vaccines.

University of Illinois professor Olgica Milenkovic believes that it’s possible to create more consistently effective vaccines -- and that the solution is better math.

Milenkovic and her research team have developed a new mathematical model for predicting which viral strains will be the most virulent and harmful to the general population. Vaccine makers can use the information to create more effective vaccines for influenza and other viruses.

“When you design a vaccine, you usually try to target three or four strains of influenza,” said Milenkovic, an assistant professor of electrical and computer engineering and a resident researcher in the Coordinated Science Laboratory. “What our model does is help you choose the three or four strains most likely to do the most harm.”

Currently, researchers at the Center for Disease Control and Diagnostics use random, one-dimensional sampling throughout the country to identify the viruses that appear most frequently and seem to have the largest probability of spreading. Viruses are usually classified based on their coat proteins, using sequence alignment or phylogenetic approaches. The center also works with foreign governments since travelers often bring new strains into the country.

But this approach has limitations. Viruses and their DNA mutate, and mutations accumulate with time. Furthermore, sometimes members of an entire household fall ill at the same time, and this throws off the numbers used for virulence prediction. If people live together, it’s more likely that a virus will spread, but it’s not necessarily the best indicator of the virulence of each strain. A home may have 10 people who all catch a mild virus, but the results should be weighed (and perhaps discounted) against a household with four people who get violently ill with an aggressive strain.

Milenkovic’s probability model tackles these problems with algorithms that account for spatial and time coordinates related to sampling, using very few sample observations. The model reveals the correct distribution of viruses, the frequency of viruses in the population even using a small sample, the percentage of viruses that may be missing from the sampling, and the original frequencies of the viruses independent of sampling techniques. The model represents a generalization of techniques used at Bletchley Park during the second World War for breaking the Enigma code, and was originally designed by Alan Turing.

“Even when you don’t know some crucial parameters, the algorithm still works for the most relevant cases,” she said.

The algorithm easily detects the most aggressive viruses. The group also developed an iterative algorithm that identifies viruses that aren’t as virulent, but spread easily, when compared with 200 virus sequences. Milenkovic’s team is still testing the algorithm against larger populations.

In addition to the flu, the model can be applied to many other viruses such as smallpox. The model does not work against retroviruses such as HIV, since their mutation rates are extremely high.