MemoCell - Bayesian inference of stochastic cellular processes with and without memory in Python.
MemoCell requires an installation of a recent Python version; Python can be installed via Anaconda.
Make sure to have graphviz and pygraphviz installed before installing memocell; for
example by executing the following in the terminal:
conda install graphviz conda install pygraphviz
Then memocell can be installed by running:
pip install memocell
Other dependencies should be installed automatically during the memocell installation.
A small workflow example of MemoCell is introduced (and continued in the code examples). We look at a simple stochastic process of cell divisions, where each realisation (/experimental repeat) starts from a single cell initially (Figure below, left panel). This initial cell and all its daughter cells will divide according to a certain distribution of division times. We want to infer this waiting time distribution which is often hard to measure directly.
To do this, MemoCell requires (typically more accessible) cell count level data. In this case, these could be cell numbers observed at one single time point after the start of the experiment/process (Figure below, right panel).
With the specification of a prior model space (which we skip here, see code examples), MemoCell will update this prior by the data to obtain posterior knowledge. Afterwards, Bayesian-averaged outputs over the complete posterior model space are computed for faithful inferences, such as an estimate for the distribution of cell division times. The estimate by MemoCell recovers the unobserved, ground truth of the test data set (Figure below, left panel).
Internally, MemoCell compares data and stochastic models by summary statistics of the so-called moments (mean, variance, covariance of cell counts). This allows exact and relatively fast inferences of the (possibly non-Markovian) stochastic models.
MemoCell is designed for inferences in multi-reaction pathways of multiple cell types as well; to learn, for example, reaction rates, general phase-type waiting time distributions or model topologies (Figure above, right panel).
Of course, one may also apply MemoCell to any other discrete-state-space time-continuous data, such as gene expression and mRNA count data; requirement is that the processes of interest can be represented by the set of zero- and first-order reaction types available in MemoCell.
Concrete code and usage examples can be found in the above folder examples
(as jupyter notebooks). If they don't render correctly by github,
copy-paste the entire URL into nbviewer;
alternatively, you may download and run them yourself.
Documentation can be found at readthedocs. It includes an extensive API detailing all classes and functions of MemoCell.
This package can be used under the MIT License (MIT), see LICENSE file.
MemoCell was written and developed by Maurice Langhinrichs and Lisa Buchauer @TSB.
The release paper of MemoCell can be found here [TODO add link].
Please cite this publication as
TODO add citation
This work is based on many people's previous achievements; please find the complete list of references in our release paper.