Directed acyclic graphs provide an effective framework for learning causal relationships among variables given multivariate observations. Under pure observational data, directed acyclic graphs encoding the same conditional independencies cannot be distinguished and are collected into Markov equivalence classes. In many contexts, however, observational measurements are supplemented by interventional data that improve directed acyclic graph identifiability and enhance causal effect estimation. We propose a Bayesian framework for multivariate data partially generated after stochastic interventions. To this end, we introduce an effective prior elicitation procedure leading to a closed-form expression for the directed acyclic graph marginal likelihood and guaranteeing score equivalence among directed acyclic graphs that are Markov equivalent post intervention. Under the Gaussian setting, we show, in terms of posterior ratio consistency, that the true network will be asymptotically recovered, regardless of the specific distribution of the intervened variables and of the relative asymptotic dominance between observational and interventional measurements. We validate our theoretical results via simulation and we implement a Markov chain Monte Carlo sampler for posterior inference on the space of directed acyclic graphs on both synthetic and biological protein expression data.

Castelletti, F., Peluso, S. (2024). Bayesian learning of network structures from interventional experimental data. BIOMETRIKA, 111(1 (March 2024)), 195-214 [10.1093/biomet/asad032].

Bayesian learning of network structures from interventional experimental data

Castelletti, F
;
Peluso, S
2024

Abstract

Directed acyclic graphs provide an effective framework for learning causal relationships among variables given multivariate observations. Under pure observational data, directed acyclic graphs encoding the same conditional independencies cannot be distinguished and are collected into Markov equivalence classes. In many contexts, however, observational measurements are supplemented by interventional data that improve directed acyclic graph identifiability and enhance causal effect estimation. We propose a Bayesian framework for multivariate data partially generated after stochastic interventions. To this end, we introduce an effective prior elicitation procedure leading to a closed-form expression for the directed acyclic graph marginal likelihood and guaranteeing score equivalence among directed acyclic graphs that are Markov equivalent post intervention. Under the Gaussian setting, we show, in terms of posterior ratio consistency, that the true network will be asymptotically recovered, regardless of the specific distribution of the intervened variables and of the relative asymptotic dominance between observational and interventional measurements. We validate our theoretical results via simulation and we implement a Markov chain Monte Carlo sampler for posterior inference on the space of directed acyclic graphs on both synthetic and biological protein expression data.
Articolo in rivista - Articolo scientifico
Bayesian model selection; Causal inference; Directed acyclic graph; Intervention; Markov equivalence; Posterior ratio consistency; Structure learning;
English
11-mag-2023
2024
111
1 (March 2024)
195
214
none
Castelletti, F., Peluso, S. (2024). Bayesian learning of network structures from interventional experimental data. BIOMETRIKA, 111(1 (March 2024)), 195-214 [10.1093/biomet/asad032].
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/416302
Citazioni
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 1
Social impact