Filzbach

Inspiration | Features | Try online | Download | Case studies | People | Acknowledgements

Fit complex models to hetereogenous data: Bayesian and likelihood analysis made easy

Filzbach is a flexible, fast, robust, parameter estimation engine that allows you to parameterize arbitrary, non-linear models, of the kind that are necessary in biological sciences, against multiple, heterogeneous data sets. Filzbach allows for Bayesian parameter estimation, maximum likelihood analysis, priors, latents, hierarchies, error propagation and model selection, from just a few lines of code.

Inspiration

Traditionally, ecology and biology has been largely split into the purely empirical (generating or analysing data with only informal use of models) and the purely theoretical (analysing models that have been, at best, only informally constrained with data). However, to create a precise, predictive, understanding of ecological and biological systems it is necessary to bridge this gap, using data to formally parameterize, and select between, competing models.

Features

• Specify parameters, define the likelihood, and Filzbach does the rest
• The automatic adaptive MCMC sampling algorithm copes with a wide range of different problems with no need for any manual tuning
• Automatic handling of multiple chains, testing for convergence, calculation of MLEs, posterior means and credible intervals on all parameters; and AIC, BIC, DIC 
• Easy error propagation of parameter uncertainty through any model
• Fast and robust compared to commonly used alternatives
• Comes with a library of easy to use parameter distributions -- but can be extended include any others, so long as they are written in C 

Try online

Get a feel for Filzbach by trying our web sampler which features the following pre-defined illustrative models:

• Simple statistical models like linear or logistic regression.
• Biologically interesting models like species distribution or tree mortality.

Download

To code up and run your own Filzbach analysis in C++, C# or another programming language download the Filzbach package which includes all necessary libraries, many examples, and a user guide.

Case studies

• Matthew Smith and Drew Purves at Microsoft Research combined FetchClimate and Filzbach to create a new model of the global carbon cycle
• Greg McInerny and Drew Purves used Filzbach to correct for fine-scale environmental heterogeneity in species distribution modelling (SDM). Read the paper or listen to the podcast
• Neil Dalchau used Filzbach to parameterize a model of MHC Class I presentation in the human immune system

People

Filzbach was written by Drew Purves and Vassily Lyutsarev at Microsoft Research, Cambridge.

Acknowledgements

Many collaborators and students have used Filzbach, provided helpful feedback and reported issues. Andreas Heil worked on the Filzbach probability distributions of the implementation of parallel chains.