Contact info: thomas.sauter@uni.lu or sebastien.delandtsheer@uni.lu or seb@delandtsheer.com
Please cite :
De Landtsheer et al., FALCON: A Toolbox for the Fast Contextualisation of Logical Networks (2017), Bioinformatics Link
De Landtsheer et al., Using Regularization to Infer Cell Line Specificity in Logical Network Models of Signaling Pathways (2018), Front. Physiol. Link
Del Mistro & Lucarelli et al., Systemic network analysis identifies XIAP and IκBα as potential drug targets in TRAIL resistant BRAF mutated melanoma (2018), npj Systems Biology and Applications Link
Follow us on Twitter : FALCON toolbox
Copyright: This software is freely available for non-commercial users under the GPLv3 license.
This updated version, released in July 2020, refactors the functionning of the toolbox and provides several bug fixes and improvements. Most notably, we removed several variables from the main functions that became irrelevant, improved the plots to make them more readable and intuitive, and integrated the various regularization schemes. We really hope these changes will improve the usability and usefulness of the toolbox and we encourage users to contribute to the project, and/or post their positive or negative experiences on the repository page.
Here is a list of the main changes in v1.3:
- Simplified API includes calls to the main functions with standard R = f(X, y) syntax.
- Parallel computations becomes the default behavior.
- The model single variable ('estim') contains the complete information for the optimization problem, as well as the results of the different analyses.
- Most hyperparameters are now set by default and user-modified if necessary only. The steady-state threshold and the maximum number of runs are automatically determined by the size of the model.
- Weighted optimization allows to weight the datapoints for the MSE calculation by an arbitrary value to give more importance to some datapoints. This is useful for model exploration.
- Additional 'likelihood' option to weight the datapoints by the inverse of their variance, therefore performing a maximum-likelihood computation. When no error is reported a default value of 5% is used.
- The 'HLBound' option is not supported anymore. Network definition files with this column are still read but all parameters are estimated on the full [0 1] range.
- The calculation of errors and information criteria during virtual knock-outs has been changed. The measurements corresponding to a knock-out node are not accounted for in the calculation anymore, in such a way that the error due to comparing the node value (zero by definition) and the measurements is eliminated. While this calculation is technically not accurate, we believe it is more useful when comparing the effect of different knock-outs.
Here is a list of the main changes in v1.2:
- The 'Forced' variable has been removed and the related quantity fixed to 1. This makes the framework completely coherent from the mathematical point of view, as all nodes now are 'forced' to abide the rules of probability.
- The option to set 'high' and 'low' parameters, i.e. to bias the constrains in favor of some interactions based on prior knowledge becomes optional. We hope this makes the model input format clearer.
- Network model files and data files in the '.csv' format are now accepted. We hope this solves in an effective way issues related with the use of Excel on Mac, Linux, or older versions of Matlab. Example input files have been updated accordingly. As csv is the de facto format for data science flat files, we might deprecate the '.txt' and Excel formats in later versions.
- Network models can be loaded in the '.sif' format. This makes FALCON model files readable by other software, for example CytoScape. Parameter names will be infered from the nodes names. All interactions will be considered linear (no Boolean gates).
- Data files now need to contain a first column termed 'annotation', which is a string (for example 'EGF+NoInhibitor-24h') associated with each condition, used for identifying this particular condition in later analyses and plots. Example input files have been updated accordingly.
- Each instance of FALCON will now generate a log file with basic info about data, model, hyperparameters and runtime info.
- The optimized metric is now MSE (Mean Squared Error). It is equal to the SSE (Sum of Squared Errors, used in the previous version) divided by the number of datapoints, and has the advantage of being a more sensible (not fool-proof though) measure to compare model fitness between different models and datasets.
- We spotted a bug in the use of Mersenne-Twister pseudo-random number generator, which seed is now really randomized according to the system status.
- The function FalconKONodes has been expanded to include the option of partial knock-downs, in order to assess the effect of these on every other network node in the different conditions. It is meant to predict the effect of siRNA experiments, which rarely achieve 100% silencing.
- We introduce the new function FalconPreProcess to handle data prior to optimization. Options include undersampling, oversampling, noise addition, normalization, bootstrapping and randomization along samples and/or conditions.
- The objective function FalconObjFun now returns the MSE, SSE, AIC, BIC, and the number of parameters in the model (which could be different than the starting number in case of regularization).
- Many regularization schemes have been added: Straight: applying a partial norm to prune a network; Grouped: applying L1 norm on intra-group parameter variability across contexts to detect context-specific regulation; Smooth: applying L1 norm accross timepoints to detect timepoints where network re-wiring occurs; Uniformity: using local density disparity in the parameter space to cluster parameter together (see the related paper De Landtsheer et. al. (2018), Frontiers in Physiology, xxx). We also allow some combinations of several regularizations on the same problem. See the related documentation (Regularizations.md) for the use of regularized FALCON objective functions.
- In general, the toolbox has been brought to the field standard in term of annotation and documentation of open-source software.
FALCON is a toolbox for the efficient contextualization of logical network models. Contextualized network models provide important qualitative and quantitative information about the system being modelled. Specifically, FALCON is well suited to assess the relative contributions of different signal transduction mechanisms to the behavior of the system at steady-state. The logical formulation of the interaction between different molecules is intuitive, and often is the only available information to systems biologists. A Bayesian interpretation of the logical ‘gates’ allows for an algebraic formulation of the system and an efficient calculation of the long-term steady-state of the system given the specified inputs. A gradient-descent optimizer (fmincon) is used to minimize the error function calculated as the sum of squared residuals between the simulated and experimentally measured nodes of interest. FALCON offers multiple types of analysis including identifiability analysis, systematic knock-out and differential regulation which can be applied to various biological applications.
- Toy model
- PDGF (Trairatphisan,P. et al. (2016))
- MAPK (Saez-Rodriguez,J. et al. (2009))
- Apoptosis (Schlatter,R. et al. (2009))
In order to use the FALCON toolbox, the necessary scripts need to be included in the Matlab path. This can be done automatically by typing “FalconInstall” on the Matlab command window. The pipeline was developed under Matlab R2014b and has been successfully tested under most versions up to R2019b. It requires the build-in Matlab solver “fmincon”, which is a part of the Optimization Toolbox (http://nl.mathworks.com/products/optimization/), during the optimisation process. The parallel computing option in the FALCON pipeline which uses the “parfor” function requires the Parallel Computing Toolbox (http://nl.mathworks.com/help/distcomp/) and the plotting of network structure with optimised parameters requires the “biograph” function integrated in the Bioinformatics toolbox (http://nl.mathworks.com/help/bioinfo/) to be present.
In case you encounter any problems while running the FALCON pipeline on earlier of later versions of Matlab, please do not hesitate to contact the authors or post an issue under: https://github.com/sysbiolux/FALCON/issues.
The FALCON pipeline was rigorously tested before its release into the modelling community. Nevertheless, users might still encounter a few issues in the FALCON pipeline and we provide some potential solutions in this section.
- The FALCON scripts and examples are not added to the path
- The FALCON pipeline does not accept certain logical gate combinations
- The FALCON pipeline crashed during the analytical process
- Lommel,M.J. et al. (2016) L-plastin Ser5 phosphorylation in breast cancer cells and in vitro is mediated by RSK downstream of the ERK/MAPK pathway. FASEB J., 30, 1218–33. Link
- Saez-Rodriguez,J. et al. (2009) Discrete logic modelling as a means to link protein signalling networks with functional analysis of mammalian signal transduction. Mol. Syst. Biol., 5, 331. Link
- Schlatter,R. et al. (2009) ON/OFF and beyond--a boolean model of apoptosis. PLoS Comput. Biol., 5, e1000595. Link
- Trairatphisan,P. et al. (2016) A Probabilistic Boolean Network Approach for the Analysis of Cancer-Specific Signalling: A Case Study of Deregulated PDGF Signalling in GIST. PLoS One, 11, e0156223. Link
- Trairatphisan,P. et al. (2014) optPBN: an optimisation toolbox for probabilistic Boolean networks. PLoS One, 9, e98001. Link