Add Bertozzi_PNAS2020 (Bertozzi et al., 2020)

SIR model trained on 10 days of New York and California data from the COVID-19 pandemic (March 2020).

The published model is defined in the paper as two versions: with and without a reparameterization. The implementation in this collection has model dynamics defined in terms of a standard SIR model, but $\beta$ is calculated from the parameters in the PEtab parameter table as $\beta = \frac{R_0 \gamma}{N}$.

Table 1 in the original publication suggests that 10 days of data were used for training, which matches the implementation in this collection.

The model is generally easy to train, to produce a fit of similar quality to the paper. 1000 starts with SciPy L-BFGS-B were use to produce the optimized parameter vector (nominal values in the parameter table) and corresponding simulations.tsv. However, even 10 starts will produce similarly good fits.

The paper uses difference equations for their simulation [1] of the ODE system, which may produce differences compared to ODE solvers typically used with SBML/PEtab.

Updated in #222

[1] https://github.com/gomohler/pnas2020/blob/master/sir.R

Benchmarking-Initiative / Benchmark-Models-PEtab

Add Bertozzi_PNAS2020 (Bertozzi et al., 2020) #99