13h00 - 13h30 – Olivier Gandrillon (LBMC / ENS Lyon)
A probabilistic dynamical framework for Gene Regulatory Network inference
Joined work with Matteo Bouvier, Alexey Koshkin, Fabien Crauste, Arnaud Bonnaffoux and Olivier Gandrillon.
In this talk, I will first recall our proposal for a GRN model that is
simultaneously probabilistic, dynamical, and executable (Herbach et al.
2017; Bonnaffoux et al. 2019). It is specifically designed to reproduce
and to predict the time-dependent evolution of the gene expression
distributions that we observe at the single-cell level, for example
during a differentiation process.
I will then address two open questions: (i) How do we compare a model's output to experimental data, that is single-cell-based gene expression distributions? and (ii) How do we compare the output of two different models? We will show that the main difficulty comes from the probabilistic nature of the model: two runs of the same model with the exact same parameter values will generate two different distributions.
I will present our current proposal and argue that there is no definitive answer to those questions and that more dedicated research is needed to answer those.
Herbach, U., Bonnaffoux, A., Espinasse, T., and Gandrillon, O. (2017). Inferring gene regulatory networks from single-cell data: a mechanistic approach. BMC Systems Biology 11, 105.
Bonnaffoux, A., Herbach, U., Richard, A., Guillemin, A., Gonin-Giraud, S., Gros, P.-A., and Gandrillon, O. (2019). WASABI: a dynamic iterative framework for gene regulatory network inference. BMC Bioinformatics 20, 220.
13h30 - 14h00 – Aurélien Naldi (Lifeware / Inria Saclay)
Kinetic assumptions in Boolean networks: a case for buffering
Boolean networks are widely used to study complex biological systems, especially in absence of precise kinetic information. Their asynchronous interpretation has long been considered as more realistic than the synchronous one, as it removes some implicit kinetic assumptions. The "most permissive" semantics, which has been recently introduced, removes all known remaining assumptions and offers surprisingly good computational properties. However, this semantics also enables some dynamical behaviors which may conflict with the expected biological meaning of many models, in particular it can depend on hidden dual interactions between components of the network. We propose buffered network as a balance between the implicit kinetic assumptions of the asynchronous interpretation and the strong generalization of the most permissive semantics. Using this approach to refine the results obtained with the most permissive semantics, we identified some key analytical results which remain valid when we preserve the signs of all interactions, while others should be treated more carefully.
Dernière modification le 05/02/2021