Infectio

Agent-based modeling for Virological Plaque.

The image above shows the output of a simulation run. (top-left) shows spatial positions of infected (green) and susceptible (blue) cells, (top-right) shows concentrations of particles and molecules produced by the system in the same spatial space as top-left, and (three bottom rows) show metrics produced by the experiment, which are the radius of the plaque, average radial velocity, and susceptible and infected population sizes, from top to bottom. These metrics, possibly with others, are used to evaluate simulation with a reference which is obtained from real world plaque experiements.

How to use

Install dependecies

Conda

conda create env -f environment.yml.

venv

Could not make conda nor poetry work on hemera, using venv instead:

python -m venv ./venv;
source ./venv/bin/activate;
pip install -r requirements.txt;

Run simulation

VACV + Epithelial model

Create a config file (e.g. similiar to default config) and modify accordingly. You can also use arguments in CLI instead (take a look at --help for available options). Then with python environment active, run: python(3) infectio/models/vacv/run.py -c path/to/config.py [optional arguments].

Create a new model

TBD

Notes

• Inconsistency for Radius information; in std output, the radius to center of infection is shown. In graph plots, the center to convex center is shown.

• Default num_cells = 2750 number used in configs are based on the density of cells computed on a single image with pixel dimensions that could possibly be wrong. So be cautious about its correctness. The calculations you can find at MacBookM1->workspace/Cellpose/01_analyze_results.ipynb

• vmin and vmax of colorbar are based on 5 and 95 % of the data. If vmax is max, then the rest of the plot except for a few locations is almost in the very low range of colorbar range and not very clear. Needs better clarification, but most probably because molecule diffusion very very slow in the medium and the single point peak stays large during the diffusion for few steps.

• Assumption in code: x,y coords start from lower left. particle variable u also same, i.e. rows correspond with x and columns with y, first row is x=0 and first column is y=0. Therefore for ploting u, need to do plt.imshow(u.T, origin='lower')

• cells are points. So molecules are produced at a single location. They also do not collide, ...

• Can we avoid discretization of particle scalar fields and work completely with well defined function. For example solve the diffusion with it, and also for gradient computations be able to use this python package: numdifftools.

• More accurate gradients? Look here Structure tensor and diagonal gradients