Bicycle: Intervention-Based Causal Discovery with Cycles

Martin Rohbeck, Brian Clarke, Katharina Mikulik, Alexandra Pettet, Oliver Stegle, Kai Ueltzhöffer
Proceedings of the Third Conference on Causal Learning and Reasoning, PMLR 236:209-242, 2024.

Abstract

While a growing number of algorithms for causal discovery of directed acyclic graphs from observational and interventional data have been proposed, the robust identification of cyclic causal graphs in particular remains an open problem. Solutions to this challenge would have a considerable impact in various application domains, including single-cell genomics, where gene regulatory networks are known to contain feedback loops. Recent work has shown promise to address this challenge by describing the expression states in a population of cells as the steady-state solution of a stochastic dynamical system. However, this formulation cannot account for information on interventions in the population, and consequently, it ignores the associated causal inductive biases, which are key assets to obtain meaningful results and improve identifiability. In this work, we propose a novel method, Bicycle, which (i) infers cyclic causal relationships from i.i.d. data, (ii) explicitly accounts for information on the perturbation state of cells by a realization of the independent causal mechanism principle and (iii) models causal effects in a latent space rather than on observed data. We benchmark Bicycle in the context of existing approaches, demonstrating improved recovery of simulated causal graphs and improved out-of-distribution prediction performance on unseen perturbations in real single-cell datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v236-rohbeck24a, title = {Bicycle: Intervention-Based Causal Discovery with Cycles}, author = {Rohbeck, Martin and Clarke, Brian and Mikulik, Katharina and Pettet, Alexandra and Stegle, Oliver and Ueltzh\"offer, Kai}, booktitle = {Proceedings of the Third Conference on Causal Learning and Reasoning}, pages = {209--242}, year = {2024}, editor = {Locatello, Francesco and Didelez, Vanessa}, volume = {236}, series = {Proceedings of Machine Learning Research}, month = {01--03 Apr}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v236/rohbeck24a/rohbeck24a.pdf}, url = {https://proceedings.mlr.press/v236/rohbeck24a.html}, abstract = {While a growing number of algorithms for causal discovery of directed acyclic graphs from observational and interventional data have been proposed, the robust identification of cyclic causal graphs in particular remains an open problem. Solutions to this challenge would have a considerable impact in various application domains, including single-cell genomics, where gene regulatory networks are known to contain feedback loops. Recent work has shown promise to address this challenge by describing the expression states in a population of cells as the steady-state solution of a stochastic dynamical system. However, this formulation cannot account for information on interventions in the population, and consequently, it ignores the associated causal inductive biases, which are key assets to obtain meaningful results and improve identifiability. In this work, we propose a novel method, Bicycle, which (i) infers cyclic causal relationships from i.i.d. data, (ii) explicitly accounts for information on the perturbation state of cells by a realization of the independent causal mechanism principle and (iii) models causal effects in a latent space rather than on observed data. We benchmark Bicycle in the context of existing approaches, demonstrating improved recovery of simulated causal graphs and improved out-of-distribution prediction performance on unseen perturbations in real single-cell datasets.} }
Endnote
%0 Conference Paper %T Bicycle: Intervention-Based Causal Discovery with Cycles %A Martin Rohbeck %A Brian Clarke %A Katharina Mikulik %A Alexandra Pettet %A Oliver Stegle %A Kai Ueltzhöffer %B Proceedings of the Third Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2024 %E Francesco Locatello %E Vanessa Didelez %F pmlr-v236-rohbeck24a %I PMLR %P 209--242 %U https://proceedings.mlr.press/v236/rohbeck24a.html %V 236 %X While a growing number of algorithms for causal discovery of directed acyclic graphs from observational and interventional data have been proposed, the robust identification of cyclic causal graphs in particular remains an open problem. Solutions to this challenge would have a considerable impact in various application domains, including single-cell genomics, where gene regulatory networks are known to contain feedback loops. Recent work has shown promise to address this challenge by describing the expression states in a population of cells as the steady-state solution of a stochastic dynamical system. However, this formulation cannot account for information on interventions in the population, and consequently, it ignores the associated causal inductive biases, which are key assets to obtain meaningful results and improve identifiability. In this work, we propose a novel method, Bicycle, which (i) infers cyclic causal relationships from i.i.d. data, (ii) explicitly accounts for information on the perturbation state of cells by a realization of the independent causal mechanism principle and (iii) models causal effects in a latent space rather than on observed data. We benchmark Bicycle in the context of existing approaches, demonstrating improved recovery of simulated causal graphs and improved out-of-distribution prediction performance on unseen perturbations in real single-cell datasets.
APA
Rohbeck, M., Clarke, B., Mikulik, K., Pettet, A., Stegle, O. & Ueltzhöffer, K.. (2024). Bicycle: Intervention-Based Causal Discovery with Cycles. Proceedings of the Third Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 236:209-242 Available from https://proceedings.mlr.press/v236/rohbeck24a.html.

Related Material