Overview

Overview

There are fundamentally 3 parts to the package. If you want to simply simulate some biological circuits and maybe do parameter sweeps, you only need to know about parts 1 and 2. If you want to do parameter inference as well, then you'll need all 3 parts. If you have your own preferred method for parameter inference, you can still use parts 1 and 2 and build your own inference method on top of them.

1. The first is a reproducible and standardized support for modeling species, reactions, and reaction rates in a easily modifiable way using an API. This way, if you specify a model and want to tweak it later to get different behavior or better fits, you can just quickly edit the model using the API and proceed rather than re-write a ton of code/edit SBML directly.

2. The second is a set of fast simulators written in Cython that can be used to simulate the model. These simulators are flexible supporting many different chemical reaction models and are much faster than other packages where the simulation is done in pure Python. There are deterministic and stochastic simulators which also support delays. Additionally, Bioscrape allows for the simulation of growing and dividing cells via its cell lineage simulators. However, deterministic delayed reactions are not yet supported when simulating cell lineages (stochastic delay reactions are). It is also possible to write your own simulator, propensities, delays, or cell lineage model that easily integrate with bioSCRAPE.

3. The third piece is a set of methods for Bayesian parameter inference that use the fast simulators underneath as well as existing Markov chain Monte Carlo libraries to do parameter inference given data. In this scenario, you specify a model, you specify the parameters you would like to identify, and you specify the data that you have measured. The inference method then runs for a while (could be a loooong time) and spits out parameter estimates as well as distributions. Bayesian inference sampler using Python emcee is built-in with Bioscrape that allows for estimation of parameter distributions from time-series data. Multiple data conditions, multiple initial conditions, time points, and many other common experimental situations are already addressed with helpful bioscrape inference functions. A deterministic parameter inference interface using Python LMFit is also built-in with Bioscrape inference. The deterministic inference solves a nonlinear optimization problem to find best estimates of parameters given a data trajectory. Other than built-in inference, Bioscrape allows for easy implementation of new parameter inference that use different cost functions or apply parameter identification from different kinds of data (such as distributional data, steady-state data etc.) Refer to the inference documentation for more information.

In addition to all of these, Bioscrape lends itself to easy extension of analysis tools. For example, users may run sensitivity analysis on Bioscrape models.