Entry for:The Bioinformatics Peer Prize
Computational models are frequently used in the area of biomedicine to interpret, describe or predict dynamic profiles associated to disease progression or drug effects. These approaches could be useful for supporting the identification of targets, biomarkers and patient subpopulations with differential response to drugs. However, the use of traditional quantitative models based on differential equations is not justifiable in a sparse data situation. Furthermore, literature on complex diseases is abundant but not always quantitative. Many molecular pathways are qualitatively well described but this information cannot be used in traditional quantitative mathematical models employed in drug development. Boolean networks models are less demanding on the required data to be implemented and can provide insights into the dynamics of biological networks.
Boolean network models represent the simplest discrete dynamic models. They only assume two discrete states for the nodes of a network, ON (active) or OFF (not active), corresponding to the logic values 1 or 0. The state of each node is determined by the state of its regulator nodes (nodes that control its activation/inhibition) based on transition rules known as the Boolean functions (BFs). The main operators of Boolean dynamics are the conjunction AND, the disjunction OR and the negation NOT. Depending on the output of the BF, the state of a node can transit from one value to another as the simulation algorithm moves from an iteration to the next.
Starting from an initial condition, Boolean models eventually evolve into a limited set of stable states known as attractors. Once the model has settled onto an attractor, it will remain there for the rest of the simulation, so we focus our work in the search of these stable states.
This methodology allows the integration of all the qualitative available knowledge in the literature into a single framework to evaluate the behavior of the system under different conditions and test hypotheses about unknown aspects of the disease.
In this work, we propose a methodology to perform Boolean modeling of Systems Biology/Pharmacology networks. For that reason, we developed a simulation algorithm called SPIDDOR to calculate the evolution of the network states taking into account deterministic or stochastic strategies for the dynamic updating of the nodes.
SPIDDOR combines R and C++ languages to enable for an efficient Boolean analysis facilitating (i) model implementation and visualization, (ii) simulation of activation profiles of the components of the network, (iii) attractor analysis and (IV) a system perturbation and sensitivity analysis.
A perturbation analysis of the network coupled with clustering analysis showed potential to identify drug targets, optimal combinatorial regimens and subpopulations of responders and non-responders to drug treatment. We propose this approach as a first step towards the development of more quantitative platforms to address the current challenges in drug development for complex diseases.
Although Boolean networks cannot be used for precise estimations of molecular concentrations or drug dosing, they are useful to gain insight into the qualitative behavior of a system under study. This is especially relevant for large scale systems in which a detailed kinetic characterization of the system is not feasible due to data restrictions or limited knowledge. We consider that the methodology presented in this work can potentiate the use of Boolean networks in Systems Biology/Pharmacology by introducing a versatile tools to enrich the analysis of these systems.
5. Future ideas/collaborators needed to further research?
We build the Boolean networks manually by reading hundreds of research articles and then we introduced the resulting Boolean functions in R (in a text file) in order to perfom the simulations and analysis of these networks with SPIDDOR. Manually building a biological network may be time-consuming and (inevitably) subjective as BFs are established following the researcher criteria. Some tools exist to infer networks automatically from experimental data but different algorithms lead to different networks while different networks are generally deduced from different datasets, therefore, it is also subjective which algorithm and dataset to use. We are aware that further work needs to be done in this aspect to evaluate which methodology leads to better results in the shortest possible time.