cr_mech_coli

This package is centered around a mechanical model for rod-shaped bacteria such as E.Coli.

Model 

The model is mechanistic and coarse-grained enough such that we can estimate its parameters from even a small dataset. Its low computational cost mean that it can be readily applied in the construction of other Agent-Based Simulations within cellular_raza [1].

Fitting Methods 

To quantify if calculated results from our model match with given experimental data, we need to extract agent information from masks and compare them with calculated results. To capture the spatial effects and make our methods applicable in a wider context, we focus on the masks instead of other possible approaches. We use methods from sciki-image [2].

Image Generation 

We use pyvista [3] to create a 3D image which contains complex lighting and occlusion effects which would be harder to reproduce in a 2D plot. Afterwards, defects are inserted with opencv[4] to mimick effects such as optical aberration, sensor noise or other imperfections which can be introduced by the measurement device.

Cell Segmentation 

The data generated by such simulations can be used to create labelled images which can then be used to train deep-learning image-segmentation algorithms. The quality of training data is dependent on our ability to reverse-engineer realistic microscopy images from the generated simulation data. We are planning to use pytorch’s ‘Faster R-CNN’ [5] to tackle this task.

The images below show the steps outlined above.