Multistrain models

How to interpret changing trends?

Should we see antigenic change? Turnover? Coexistence?

Resistance to vaccines? Antimicrobials?

Model types are fuzzy


Mechanistic (mathematical v. agent-based)

Advantages and disadvantages of each?

History-based models

Track individuals' histories of infection, $S_{(0)}, S_{(1)}, S_{(1,2)},...$

Cross-immunity reduces probability of infection or infectiousness

Numerically unwieldy: $2^n$ histories for $n$ strains

Dimensional reduction

Requires cross-immunity through reduced transmission (only)

Cross-immunity must be in context of antigenic "neighborhoods"

(with $n$ strains in $m$ neighborhoods, system is $n$ x $(m+1)$)

Impossible to calculate immunity to new strains

Status-based models

Track individuals S, I, or R to each strain

$O(n)$ equations possible

"Polarized" immunity: If $i$ and $j$ cross-react, infection with $j$ leads to perfect immunity to $i$ in some fraction of hosts

A "simple" status-based model

Implicit assumption

Proportion of population not infected by second strain

Some options for multistrain models

Classic problem: invasion

Subtle differences change dynamics


Know your assumptions

Which model promotes coexistence?

Model A

Model B

Does this model bias for coexistence?



Ecological dynamics depend on strain composition

Multiplicity of infection


Ecological dynamics

Ancestor tracing

The question of strain space

Continuous or discrete?

How many dimensions?

Continuous space

Discrete space

Immunodominance and breadth affect diversity

Agent-based models

Potentially most efficient with many strains

Support complex interactions

Naturally incorporate demographic stochasticity

Analytic approaches

Equilibria: Solve for $S_{i}'=I_{i}'=...=0$

Stability analysis: Linearize at equilibria

Invasion analysis: Is $I_{2}'>0$ at $I_{1}^*$?

Integration: Calculate $I_{1}(t), I_{2}(t),...$

Numerical integration

Approximate solution to $N'=f(N)$:

$$N(t_0+\Delta t) \approx N_0+ f(N_0) \Delta t = N_1$$

Re-iterating, we get

$$N_{z+1} \approx N_{z}+f(N_{z}) \Delta t$$

$$N_{z+1} \approx N_{z}+f(N_{z}) \Delta t$$

We could assume $f(x)$ is constant for $\Delta t$ (Euler's method)

Even better, average over multiple points!

A second-order method

$$N_{z+1} \approx N_{z}+f(N_{z}) \Delta t$$

$$M_{z+1}=N_z+f(N_{z}) \Delta t$$

$$N_{z+1}=N_z+\frac{1}{2}\left[ f(N_{z})+f(M_{z+1})\right] \Delta t$$

4th-order Runge-Kutta

$$k_1=f(N_z) \Delta t$$

$$k_2=f\left(N_z+\frac{1}{2}k_1 \right) \Delta t$$

$$k_3=f\left(N_z+\frac{1}{2}k_2 \right) \Delta t$$

$$k_4=f\left(N_z+k_3 \right) \Delta t$$


Why not just pick an extremely small $\Delta t$?

Considerations for multistrain models

Form of immunity (duration, strength, polarity)

Effect of immunity (susceptibility, infectiousness, clearance)

Coinfections and timing of interactions

Accuracy for individuals v. populations

Dimensionality and discretization of strain space

Analytic and numeric approaches, including stochasticity