MEmilio v1.3.0 - A high performance Modular EpideMIcs simuLatIOn software

MEmilio implements various models for infectious disease dynamics, from simple compartmental (ODE) models through Integro-Differential equation-based (IDE) models (sometimes also denoted "age of infection models") to agent- or individual-based models (ABMs). Its modular design allows the combination of different models with different mobility patterns. Through efficient implementation and parallelization, MEmilio brings cutting edge and compute intensive epidemiological models to a large scale, enabling a precise and high-resolution spatiotemporal infectious disease dynamics.

v1.3.0 Changes

Added features / functionality:

• Allow output of mobility data for Graph-ODE model
• Added age resolution for LCT model
• Implementation of Generalized Linear Chain Trick model
• Allow variable dynamic NPI delay implementation
• Update of Damping now allows negative coefficients (i.e., contact increases)
• Added serialization to ABM
• Added functions for a person in the ABM to choose whether to comply to mask,test and isolation
• Allow explicit stepper wrappers (i.e., fixed step size numerical integrators)
• Bind different numerical integrators for python
• Allow download of population data for different years
• Plot functions for AST in generation package

General changes:

• Redesign of python bindings structure to improve typing
• Reduction of export_input_data_county_timeseries function
• Reduce use of get_support_max method in IDE model to reduce run time
• Handle Epidata FutureWarnings

Corrections:

• Error when compiling on Mac with new boost
• Missing includes in python bindings
• Deleted incorrect todo in ABM code
• divNj in ODE models creates NaN values if subpopulation is zero
• Get_default of parameter HighViralLoadProtectionFactor did not work as expected

v1.2.1 Changes

Added features / functionality:

• Implementation of stochastic two variant SEIR models
• Added delay in testing and planned mobility in ABM
• ScoreP performance profile of ABM
• gperftools profiler
• New initialization method from flows for LCT model
• Stub generation for python bindings

General changes:

• Refactored ABM code: Rename EPI to MIO, migration to mobility, and World to Model
• Replaced pointers in ABM by new structure
• Make the LCT secir model a derived class of CompartmentalModel
• New parameters that describe the multiplicative factor used for the TestAndTraceCapacity
• Extended test for equilibrium of IDE model
• Change return type of check_constraints function in the populations class to bool
• Adapt parameters_io of IDE SECIR model
• Small refacturing of epidata package
• Remove necessity for credentials of population data download
• Download boost libs with URL instead of git repo
• Upgrade manylinux image used for python in CI
• Workaround for github action checkout@v3

Corrections:

• Add warning for end_date later than 2024-07-21 in intensive care data
• OdeIntegrator could run indefinitely
• Removed bug in getting test parameters in specific test derived from generic test

v1.2.0 Changes

Added features / functionality:

• Stochastic differential equation based SIR and SEIR models
• Linear Chain Trick ODE-based model with initialization methods for real world data
• Automatic differentiation for ODE-based models and dynamic optimization examples
• Allow contact increase for simulation of larger events
• Allow flexible start day in IDE SECIR model
• Added seasonality for IDE SECIR model
• Alternative computation of compartments in IDE SECIR
• Implement initialization scheme for flows in IDE SECIR model
• Add Gamma distribution and other parameters to state age function for IDE models
• Python support for ODE SECIRVVS model
• Python support for 2021 metapopulation/Graph-ODE SECIRVVS simulation
• Age group resolution for ODE SIR and SEIR models
• Use ccache in CI for linux builds

General changes:

• Use times for exposed and infected, no symptoms state in particular ODE models instead of SerialInterval and IncubationTime
• Updated CI actions
• Updated epidata readme
• Improve IDE SECIR model readme
• Handle pandas read excel engines
• Bundle the boost git repo instead of providing a targz archive
• Streamline ODE SECIR python code

Corrections:

• Corrected handling of minimal step size in numerical integration
• Corrected functionality of IDE SECIR model example
• Prevent NaNs in newly added SDE models
• Resolve size_t underflow in dynamic NPIs
• Fix failing RKI urls
• Make python serialization working again
• Corrected IDE SECIR model simulation for certain conditions
• Corrected gcc compiler version in CI

v1.1.0 Changes

Added features / functionality:

• Graph simulation with metapopulation model for Munich
• Computation of reproduction number for ODE SECIR model
• Machine learnt surrogate model for ODE SECIR model with multiple age groups and contact change points
• Linear Chain Trick SECIR model
• New initialization for IDE model
• Unit Tests with OpenMP

Corrections:

• Correct selection of specialized simulation and advance functions in python bindings
• Corrections for new MSVC

Other:

• Expanded tests for python bindings simulations
• Small changes and fixes (logo, pull request template, ...)

In version 1.0.0, we publish:

Basic models (with local focus or without spatial resolution):

• four different ODE-based models from simple SIR to extended models with three subpopulations of different immunity levels and eight different compartments from asymptomatic to severe and critical disease states
• two IDE-based models in which more realistic transmission and compartment stays can be realized
• one agent-based model (ABM) which, due to its object-oriented implementation, allows for simulation of different immunity levels and multiple virus (variants)
  --> All models can be resolved for demographic features such as age or income.

Inflow and outflow computation for compartmental models
Basic compartmental models inherit from either a parental CompartmentalModel or a FlowModel so that new ODE-based models with standard analyses tools can be implemented time-efficient. In contrast to classical implementations of ODE-based models, FlowModels ensure a continuous computation of inflows and outflows of the compartments such that, e.g., new hospitalizations can be tracked easily.

Mobility concepts which leverage basic models to spatially resolved models

• A deterministic mobility concept with predefined round-trip trajectories.
• A stochastic mobility concept which allows for non-deterministic mobility.

Parameters and demography
Parameters and demography are implemented by generic concepts such that they can be easily extended to more general lists of parameters or additional stratifications like age or income.

Ensemble run concepts
Via standardized implementations, parameter sampling and ensemble run simulations can be conducted to assess uncertainty of the particular model outcomes.

Optimizations

• MPI-parallel implementation of ensemble runs for parameter sampling for ODE-based models
• OpenMP-parallel implementation of agent-based models
• Optimizations towards compile-time evaluation of software parts.

Helpers, utilities, math, ...
MEmilio also provides a lot of mathematical algorithms, helper tools, and utilities and to simulate or analyze results.

Tests and benchmarks
The MEmilio C++ backend is largely covered by software and unit tests (>95%) and benchmarks for some models are already available. A continuous integration pipeline ensures functionality of the software.

Python frontend to efficient C++ backend
To open MEmilio to python developers, a variety of implemented C++ models can already be called from python via the memilio-simulation package.

Python scripts for Sars-CoV-2 and demographic data
In order to run simulations for Sars-CoV-2 in Germany, several official data sources can be downloaded and postprocessed uniformly by the memilio-epidata package.

Model code generation
Due to the standardized structure of compartmental models, a part of new model code can be automatically created via the memilio-generation package.

Surrogate modeling
With the memilio-surrogatemodel package, expert models will be considered to be replaced by artificial intelligence and neural networks.

Visualization
MEmilio also already provides certain tools for visualization of simulation results.

In order to understand MEmilio, a lot of examples have already been implemented.

MEmilio quality control via detailed review processes ensures validation of implemented code concepts by independent developers.

For more details, see the readmes on https://github.com/SciCompMod/memilio in the particular (sub)directories.