developed at the Process Control & Informatics Unit

The Unit has developed the following software:

  • Jaqpot Quattro (JQ) web application for predictive model development, validation, testing and calculating model predictions. The Application Programming Interface (API) adheres to the the REST architectural constraints, is an extension of the original OpenTox API and has been documentated using swagger:

A Jaqpot Docker image is available at        that guarantees reproducibility of results and full compatibility at all times.  


  • A number of API compatible web services that implement major machine learning, and statistical algorithms used for preprocessing, regression, classification, clustering and model validation and for defining the domain of applicability. An indicative but not exhaustive list of methods include: multivariable linear regression, Lasso, Elastic net, hierarchical clustering, bi-clustering, ID3 decision tree, partial least squares, partial least squares with Variable Importance in Projection (VIP) variable selection, radial basis function neural networks, support vector machines. The Jaqpot Protocol of Data Interchange (JPDI) allows developers of modelling algorithms to integrate their implementations in the framework, using the OpenTox algorithms ontology.   JQ provides full integration with  the R and Python languages and the WEKA machine learning software, so any algorithm written in these languages can be included in the system:


  • The JQ user interface ( had its first beta release in April 2016. It enables non modelling-proficient users to have easy access to all JQ tools and applications, but also to a repository of validated models. Second release coming soon! 


           The current application can be used for particles of spherical or similar shape, but we plan on extending it for nanotubes and other particle shapes.

  • R package that performs optimal experimental design. Full factorial design and the D-optimal, A-optimal and I-optimal experimental design algorithms are included (

  • R package for training dose-response models based on the PROAST software. The model can identify the dose of a substance that is expected to result in a pre-specified level of effect  (benchmark response, BMR).

  • RRegrs package: An R package for computer aided model selection and easy-to-use model comparison (  It automates the estimation of the best regression model by using certain cross validation schemes and by searching over many different algorithms and tuning the parameters in each algorithm. Other resampling procedures are also included to minimize the variability errors, such as data splitting, repeated cross-validation, Y-randomization. Additionally normalization and filtering options are included. (Collaboration with University of Maastricht).

  • A user application for creating  and simulating PBPK models and for calculating optimal continuous dosing strategies based on the Model Predictive Control (MPC) methodology  (


  • MyDosage is a mobile friendly web application that helps doctors personalize the dosage of drugs for their patients. Personal data is used as input along with the different input requirements that each drug algorithm requires. Reports are exported in the form of graphs and values (


  • ToxFlow. A web-application developed for enrichment analysis of omics data and toxicity prediction. A sequential analysis workflow is suggested where users can filter omics data using enrichment scores and incorporate their findings into a correlation-based read-across technique for predicting a toxicity index based on its analogs. Either embedded or in-house gene signature libraries can be used for enrichment analysis. Visualization options are offered to interactively explore correlation patterns in the data, whereas results can be exported for further analysis.